Wanted to compare notes with people doing similar things.

My current setup is TensorRT-LLM with 5090, and I keep running into continuous batching.

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

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

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

This isn't revolutionary, but the combination of TensorRT-LLM and Threadripper box felt surprisingly usable once I tuned around embeddings.

Maybe useful as a starting point for newer folks.

Maybe I'm missing something obvious.

I'm mostly working with Llama in SGLang on dual 3090. The part I'm trying to understand is small local stacks.

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

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.

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 continuous batching.

Would appreciate practical advice rather than theory.

This is mostly for local work rather than public serving.

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

I've gone back and forth between SGLang and vLLM for DeepSeek. The interesting difference for me has been reranking.

Interested in tradeoffs more than fan loyalty.

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

I tried this with Kimi using llama.cpp on 7900 XTX. The headline result was better than I expected, but prefix caching still ended up being the limiting factor.

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

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.

This isn't revolutionary, but the combination of MLX and Threadripper box felt surprisingly usable once I tuned around batch size.

Maybe useful as a starting point for newer folks.

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

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

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

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

This is mostly for local work rather than public serving.

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

I've gone back and forth between PyTorch and SGLang for Mistral. The interesting difference for me has been INT4.

Interested in tradeoffs more than fan loyalty.