top | item 45837602

(no title)

enigma101 | 3 months ago

what's the hardware needed to run the trillion parameter model?

discuss

order

CamperBob2|3 months ago

It's an MoE model, so it might not be that bad. The deployment guide at https://huggingface.co/moonshotai/Kimi-K2-Thinking/blob/main... suggests that the full, unquantized model can be run at ~46 tps on a dual-CPU machine with 8× NVIDIA L20 boards.

Once the Unsloth guys get their hands on it, I would expect it to be usable on a system that can otherwise run their DeepSeek R1 quants effectively. You could keep an eye on https://old.reddit.com/r/LocalLlama for user reports.

wg0|3 months ago

Are such machines available in the A class clouds such as Azure/AWS/Google?

trvz|3 months ago

To start with, an Epyc server or Mac Studio with 512GB RAM.

criddell|3 months ago

I looked up the price of the Mac Studio: $9500. That's actually a lot less than I was expecting...

I'm guessing an Epyc machine is even less.

graeme|3 months ago

How does the mac studio load the trillion parameter model?