Issue with AI Features in Version 30.0.6 – Please Hold Off on Updating

Hey everyone,

We’ve identified a critical issue in version 30.0.6 that affects our AI features. If you rely on these features, we strongly recommend holding off on updating for now.

The good news is we already have a fix ready! We’re working on getting it released as soon as possible and will update you once it’s available. If you already updated and don’t want to wait for a new release you can apply the following patch to remedy the problem: [stable30] fix(TaskProcessing\Manager): Always use distributed cache and use PHP serialize by backportbot[bot] · Pull Request #50682 · nextcloud/server · GitHub

Thanks for your patience, and we sincerely apologize for the inconvenience!

Stay tuned for updates, and feel free to reach out if you have any questions.

7 Likes

So I guess the paths forward are …

  1. Hold off

  2. Upgrade and patch

  3. Wait for 31

  4. Wait for 30.0.7

I’m not sure if there’s a path for a new 30.0.6 or earlier 30.0.7; at least that doesn’t show up on the github pages.

Yep

Nope

Not sure, but I wouldn’t hold my breath.

Personally, I would do the following:

If you are still on 30.0.5, wait for 31 or 30.0.7.

If you have already updated to 30.0.6, and you are relying on the A.I. features, I would apply the patch manually, which is actually fairly easy to do: Patching Nextcloud — Nextcloud latest Administration Manual latest documentation

1 Like

I tried many times to make it work on 31.0.0
The chat may respond sometimes, but i can’t explain when and why it works 1% of the time… Hope next version will be more stable an reliable

Which “AI features” exactly? I got everything set up correctly. AI works. and super fast too ( added 6 workers to it ) but it has it’s weird bugs like it can NEVER generate a title ( for a conversation ) in the language that’s actually spoken in the conversation.
It will try portugese, french, russian, I think I even saw south asian languages, everything except mine.

Also in the “mail” app, sometimes it “tries” to recommend me its recommended reply (underneath the mail), but most of the time it’s very off and not usuable.

Not working here. Patched 30.0.6 and it isn’t launching the task to the GPU. Restarted the PHP and Nginx app. Even just bounced the machine itself. Still spin cycling its laundry…

Just to add some use cases here, I’ve added two additional LLMs, and no longer work even when used with the stock LLM (Health Ranger Neo gguf file and a DeepSeek gguf file, which hasn’t been attempted at 29). Chat with AI with NEO.gguf and even Whisper Large Model mp3 conversion to text (STT) would never hit the GPU as they did on 29.0.7.

Where are the error logs? Are they within the container? Other than that, how can I tell whether the taskmanager launches?

Thanks!

The AI Chat and other functionality isn’t working yet in 30.0.7. I am able to get Whisper to work, and uninstalled just in case there was some conflict. Nope.

All of the AI components have been removed and then re-added in the specific order. Four background-workers have been added into the systemd services tasks. The permissions and execution of the docker socket and services have been verified.

Nextcloud simply does not get the message, or fails to submit the execution, related to querying the Large Language Model (2).

This Whisper speech-to-text does function, however.

–Joe.

The entire AI deployment has been removed. The containers were removed when stopped, and all of the images were also removed. The Nextcloud Chat with AI isn’t operational.

Then the AppAPI was added along with a Deployment test. Nextcloud Assistant was added as an App. Then the LLM2 was added to implement AI Chat. The system is not yet querying the LLM2 container, and isn’t showing as having engaged the GPU, as seen through the nvidia-smi tool.

Nextcloud 30.0.7
nginx 1.26.2
postgresql 16.6.1
PHP fpm 8.2
Ubuntu 20.04
NVidia 12Gb A2000
128Gb

Update 30.0.8 did not offer improvement, but then there were no specific items directly related to lack of response from AI. I’ll move directly to LLM2 guys for now.

–Joe.