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: Qwen, TensorRT-LLM, Threadripper box, with a lot of attention on tail latency.

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

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

I expected the biggest difference to come from the model, but the stack around it mattered at least as much.

Wanted to compare notes with people doing similar things.

My current setup is PyTorch with mini PC, and I keep running into tokens/sec.

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

My use case is pretty ordinary: local chat, some document work, and occasional batch jobs.

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

This isn't revolutionary, but the combination of vLLM and mini PC felt surprisingly usable once I tuned around tool calling.

If someone wants the exact settings I can clean them up and post them.

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 KV cache pressure.

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

This is mostly for local work rather than public serving. I keep promising myself this is the last rebuild.

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

I've gone back and forth between SGLang and PyTorch for Qwen. The interesting difference for me has been batch size.

Interested in tradeoffs more than fan loyalty.

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

I tried this with Llama using PyTorch on mini PC. The headline result was better than I expected, but consumer GPUs still ended up being the limiting factor.

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

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

This isn't revolutionary, but the combination of SGLang and Threadripper box felt surprisingly usable once I tuned around long context.

Maybe useful as a starting point for newer folks.

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

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 INT4.

Wondering if this is just where the whole space is heading now.

I figured I'd ask here before changing everything.

My current setup is TensorRT-LLM with 7900 XTX, and I keep running into NVFP4.

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

Curious what people here are seeing.

I'm mostly working with Gemma in Ollama on 4090. The part I'm trying to understand is tail latency.

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

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