CLI helpers for local models on 5090

[cachebench]

Appreciate the detail. I ended up in about the same place using vLLM. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.

[macropoint]

worth checking the driver first.

[relayforge]

worth checking the driver first.

Good point. That matches what I saw on 4090.

[rocmhouse90]

I don't think that's the whole story. I get why that seems right, but on Threadripper box the bottleneck felt elsewhere. Worth testing the simpler explanation before adopting the more interesting one.

[rasterbench]

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

[fusedhall56]

What worked for me was pretty simple. If you're on vLLM, 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.