You're viewing a single thread.
Im curious to see what sorts of recommended minimum specs there will be for these features. It is my understanding that these sorts of models require a non negligible amount of horsepower to run in a timely manner.
At the moment I am running Nextcloud on some raspberry pis and, my gut tells me I might need a bit more oomph than that to handle this sort of real time AI prompting >_>;
The blog post states:
We build the AI Assistant using a flexible, solution-independent approach which gives you a choice between multiple large language models (LLM) and services. It can be fully hosted within your instance, processing all requests in-house, or powered by an external service.
So it sounds like you pick what works for you. I'd guess on a raspberry pi, on board processing would be both slow and poor quality, but I'll probably give it a go anyway.
Yeah sorry I was specifically referring to the on prem LLM if that wasnt clear, and how much juice running that thing takes.
Some of the other Nextcloud stuff (like that chat stuff) isn't suitable on Raspberry Pi, I expect this will be the same. It's released though, right? Might have to have a play.
You'd be surprised at how little computing power it can take, depending on the LLM.
the AI that nextcloud is offering uses openAI, sign up get a api key and add it. Your ai requests goto the cloud. (and i couldnt get it to work, constant " too many request" or a straight "failed")
The other option is the addon " local llm", you download a cutdown llm like llama2 or falcon and it runs locally. I did get thoes all installed, but it didnt work for general prompts.
Nextcloud will probably fix things over time, and the developer who made the local llm plugin will to, but right now this isnt very useful to selfhosters.
Llamaโs getting pretty damn good, check out phind.com if you havenโt yetโฆits programming better than GPT-4 supposedly!
I just asked it to write an assembly program for the Intel 8008 uprocessor, and it just knocked it out! That's not bad for a chip that was released in 1972 !
Well, Nextcloud runs like shit on a Pi WITHOUT having to do AI stuff, so.....
People should stop suggesting Nextcloud for Rpi. I really love Nextcloud but it sometimes struggles even with somewhat decent machine.
I love Nextcloud but it's just oh so painfully slow at times
That's likely a problem with your configuration. Mine was slow too until I set up redis.
Great that you don't have any problems with it, I do at times. It's not a problem with the config either (at least not the Nextcloud config) because I set it up using the Nextcloud VM script, not manually. It's not slow all the time but when it feels all the slower for it.
Nextcloud would struggle on devices with low CPU performance and slow storage speed. A Pi checks all those box. You might increase the performance a bit by running nextcloud from an external SSD but there is no fixing the Pi's low CPU performance.
I've tried running NextCloud from a system with a SATA SSD and a Core i7 using WSL.....and it still ran like shit.
Not using redis? Mine ran like shit and I almost gave up until I set up file locking and caching.
Yeah, I'm wondering the same and also figure the requirements will be pretty significant. Still, pretty happy with things like this and Home Assistant's recent work on local voice assistants.