what hardware should i buy to run LLMs locally?

naive question, i know, but i don't have a good sense of where to start. are there useful public guides for deciding? the state of the art is changing rapidly, so figured i'd ask here first...

soldier of Good. 🪖🥰 
producer × DJ. 🎧 🪩 
working on /campfire. 🫴🪵🔥
host of /downshift ⚙️ + /dww 🎧 +  /gtd 🛠️.
https://linktr.ee/0xdownshift

have you looked at ollama? it can run even on reasonably simple machines, but it is slow if cpu only. you could get a good mac studio, or even string some machines together with exolabs. guess i hae to ask - what is the budget, and what kind of model sizes are you planning to run

Applied AI, open source developer, databases, data, vibe coding (sic.), remote. 

itinerant businessperson. globalist.

Kuala Lumpur, Federal Territory of Kuala Lumpur, Malaysia

it's a huge question; i'm working on a complex app. not even sure all models will be LLMs.

the hairiest problem is taking a URL for a certain song and then finding that same song's URL on other streaming services using metadata and any other tricks i can cook up, potentially including audio analysis (a la shazam using a spectrograph and some fancy transformations). this is especially tough for dance music (my focus) because house and techno beats can be very very similar 🫠 

and then, of course, code generation (if it provides any advantage over using centralized models—i currently have no idea tbh)

i'm sure there are other applications i will need but haven't thought of yet

yeah feel free to holler here or in DMs. the fact that you also have another use helps lighten the budget load. the key is knowing what you're trying to achieve with a local llm.

awesome pointers!

i love the mac studio, but wonder if it's cost efficient $/flop (i would also use it for audio production, tho)

i'm still early in my research but will follow up when i have a better idea on model size. i don't have a firm budget just yet 

ty 🙏🏼