Did a bit of testing and thought I'd post the result here.

After a few runs on mini PC, I ended up with decent performance, but the setup only really made sense once I adjusted around HBM constraints.

Nothing scientific here, just practical notes from a real setup.

For context, I'm not chasing leaderboard numbers. I mostly want the system to feel consistent day to day.

Posting this in case it saves someone else a little time.

I'm helping a small team set up a local workflow. We don't need a giant stack, but we do care about VRAM headroom.

Simple and boring is a valid answer.

Thought I'd post a small recipe because a lot of threads ask similar things.

Just sharing a setup that ended up working reasonably well: Kimi, MLX, Threadripper box, with a lot of attention on MoE serving.

Feel free to tear it apart if you see something dumb.

For context, I'm not chasing leaderboard numbers. I mostly want the system to feel consistent day to day.

I'm trying to sanity-check my setup.

I'm trying to decide whether to keep the current stack or change course. Main question is how people here are handling token cost.

If you've tested something similar, what did you keep and what did you stop bothering with?

This is mostly for local work rather than public serving.

This keeps coming up in conversation so I figured I'd start a thread.

It feels like the conversation has shifted hard toward inference and serving details lately, and honestly that makes sense once you start dealing with tokens/sec.

Curious whether that lines up with what other people are seeing.

Spent some time on this today and wanted to compare notes.

Two paths both seem viable, but they break down differently once quant drift becomes important.

Trying to avoid rebuilding the whole thing twice.

I care more about stability than squeezing the last possible tokens/sec out of it.

Thought I'd post a small recipe because a lot of threads ask similar things.

Maybe useful as a starting point: SGLang on dual 3090, keeping KV cache pressure in mind from the start.

Maybe useful as a starting point for newer folks.

Posting this in case it saves someone else a little time.

I'm helping a small team set up a local workflow. We don't need a giant stack, but we do care about INT4.

Would rather learn from your scars than collect my own all at once.

This keeps coming up in conversation so I figured I'd start a thread.

One thing I keep noticing is that people underestimate structured output until they actually try to run this stuff for more than one afternoon.

Maybe obvious to some of you, but it clicked for me this week.

Maybe I'm missing something obvious.

I'm mostly working with Phi in LM Studio on 4090. The part I'm trying to understand is VRAM headroom.

Not looking for the perfect answer, mostly what worked in real use.