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

Two paths both seem viable, but they break down differently once NVFP4 becomes important.

Trying to avoid rebuilding the whole thing twice.

Curious what people here are seeing.

I'm mostly working with Mixtral in SGLang on A6000. The part I'm trying to understand is KV cache pressure.

Would appreciate practical advice rather than theory.

This is mostly for local work rather than public serving.

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 A6000 felt surprisingly usable once I tuned around prefill/decode split.

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

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

If I had to summarize the lesson so far, it would be that clean setup and observability beat clever guesses.

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

After a few runs on dual 3090, I ended up with decent performance, but the setup only really made sense once I adjusted around tokens/sec.

Curious if anyone else landed in the same place.

This is mostly for local work rather than public serving.

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

This isn't revolutionary, but the combination of Ollama and dual 3090 felt surprisingly usable once I tuned around quant drift.

Maybe useful as a starting point for newer folks.

This is mostly for local work rather than public serving.

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

This isn't revolutionary, but the combination of llama.cpp and dual 3090 felt surprisingly usable once I tuned around batch size.

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

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

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

I've gone back and forth between llama.cpp and TensorRT-LLM for DeepSeek. The interesting difference for me has been batch size.

Interested in tradeoffs more than fan loyalty.

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

Maybe I'm missing something obvious.

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

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

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

I'm comparing two approaches here: keep it simple on 5090, or squeeze more out of the stack and deal with the complexity.

What would you pick if you had to live with it for a few months?

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

Short version: PyTorch behaved well for me, but only after I stopped ignoring MoE serving.

Happy to post more details if anyone wants them.