synthetic data for support bots: setup advice

[cacheworks]

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

Short version: vLLM behaved well for me, but only after I stopped ignoring NVFP4.

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

[batchember]

same here on my side.

[routecraft]

same here on my side.

Maybe, but I had the opposite result. I get why that seems right, but on A6000 the bottleneck felt elsewhere. Worth testing the simpler explanation before adopting the more interesting one.

[routeline23]

Good point. I don't think your conclusion is far off. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[quantsupply]

Appreciate the detail. That matches what I saw on Threadripper box.

[quantghost]

Appreciate the detail. That matches what I saw on Threadripper box.

Did a small comparison this afternoon. On 7900 XTX with LM Studio, I saw basically the same behavior. I expected the model choice to matter more than it did.

[rocmplane]

Need a little more detail here. Are you testing this on 4090 or something else?

[mellowwire98]

A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with Phi and compare from there.

[rocmplane]

My first guess would be this: If you're on llama.cpp, I'd confirm the version first and then reduce variables until the behavior is obvious. Not glamorous advice, but repeatable baselines save a lot of time.

[routecraft]

subscribed, curious where this lands.