LM Link enabled remote access via iPhone and MacBook
For WWDC, we worked with Apple to run Kimi K2.6, a 1T-parameter model, across a cluster of four Mac Studios using a preview version of LM Studio.
We showcased secure remote access from a MacBook Neo and iPhone using LM Link.
A glimpse of your own private, frontier-scale AI.
Many users praised LM Studio for running a 1T-parameter Kimi model locally across four Mac Studios as a privacy breakthrough, while some objected to the high hardware cost and questioned the model's quality.
No Digg Deeper questions have been answered for this story yet.
@lmstudio How many tokens per second
@lmstudio Will you guys be testing the new AMD AI MAX Pro?
The future. Fable who?
For WWDC, we worked with Apple to run Kimi K2.6, a 1T-parameter model, across a cluster of four Mac Studios using a preview version of LM Studio.
We showcased secure remote access from a MacBook Neo and iPhone using LM Link.
A glimpse of your own private, frontier-scale AI.
@lmstudio Too bad they don't even sell those 516GB models anymore.
@lmstudio Arent 2 enough for this model (at q4)
@lmstudio @yagilb Where is the “add to wishlist” button on this app?
@lmstudio Just as soon as 512Gb Mac Studios are a thing again …
@lmstudio That's a glimpse only if you can afford the hardware setup.
@lmstudio running a 1T model on macs is genuinely sick but the real flex is lm link making it feel local from an iphone... latency on that preview gotta be wild tho, what's it looking like
@lmstudio @grok how much does that cost the monitor and everything else
@lmstudio Great! Now we just need the non existent MacStudio machines…
@lmstudio Neat! The four node cluster is cool but it’s going to be amazing to watch it turn into a single node and laptops and mobile devices.
@lmstudio The future favors AI hardware for off-network use.
@lmstudio Insane stuff
@lmstudio @AlicanKiraz0
@lmstudio @mattturck Question is how did you procure the 512GB RAM spec for the Mac Studio 🤷🏻♂️🤦🏻♂️
@lmstudio 4 Mac Studios running a 1T-param model offline is a pretty compelling demo of what local inference can look like when you don't need a datacenter. Curious how the token throughput scaled across the cluster vs a single H100 node.
@lmstudio i dont have 6k for mac studios but i do have a 2 old laptops with a broken screen and keyboard. Can i cluster those?
Been running LM Studio from it's inception, it's genuinely the best local inference experience on Apple silicon. The Kimi K2.6 cluster demo at WWDC is the hardware milestone that matters: a 1T parameter frontier-class model running on $16K of consumer hardware. Four Mac Studios at ~$4K each. A year ago this required a data center. Today it fits on a desk. The current pricing is real and it's not for everyone. But the trajectory is the point. Apple silicon compute per dollar has been improving faster than any other hardware category. What costs $16K today will cost $4K in two years and fit in a single machine. The direction is clear. We are moving toward a world where frontier-scale inference is personal infrastructure. LM Studio is building exactly the right software for that world.
@lmstudio @yagilb Hope you'll consider adding a user management as in the open web ui. This would be amazing if one could manage multiple accounts.
