+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: LLM Models (https://www.datasculpting.ai/inference/forumdisplay.php?fid=12)
+--- Forum: Open Weight (https://www.datasculpting.ai/inference/forumdisplay.php?fid=49)
+---- Forum: Local (https://www.datasculpting.ai/inference/forumdisplay.php?fid=51)
+---- Thread: simple local serving recipe for Kimi (/showthread.php?tid=1182)
simple local serving recipe for Kimi - mintstation99 - 10-14-2026
Wanted to compare notes with people doing similar things.
I'm trying to decide whether to keep the current stack or change course. Main question is how people here are handling HBM constraints.
Not looking for the perfect answer, mostly what worked in real use.
I expected the biggest difference to come from the model, but the stack around it mattered at least as much.
Re: simple local serving recipe for Kimi - rocmriley - 10-16-2026
What worked for me was pretty simple. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at small local stacks. Once you know the plain path works, then add the faster path back in piece by piece.
Re: simple local serving recipe for Kimi - emberpeak - 10-16-2026
A couple questions before I guess too hard. What model, what runtime, and roughly what context size?
Re: simple local serving recipe for Kimi - audionode8 - 10-16-2026
A couple questions before I guess too hard. Are you testing this on GB200 or something else?
Re: simple local serving recipe for Kimi - rasterbench - 10-16-2026
Appreciate the detail. I don't think your conclusion is far off.
Re: simple local serving recipe for Kimi - mlxgarden - 10-16-2026
That matches what I was seeing. The pattern around consumer GPUs has been pretty similar for me.
Re: simple local serving recipe for Kimi - phaseshed - 10-17-2026
worth checking the driver first.
Re: simple local serving recipe for Kimi - rocmriley - 10-17-2026
What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with Phi and compare from there. Once you know the plain path works, then add the faster path back in piece by piece.