domain tune without wrecking base behavior on B200

[benchplease]

Did a small comparison this afternoon. My result was mixed: first run looked bad, second run looked normal.

[turbosupply83]

same here on my side.

[cudascope]

I'd approach it in a boring order. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at tokens/sec. Once you know the plain path works, then add the faster path back in piece by piece.

[vrampilot]

Maybe, but I had the opposite result. I wouldn't optimize around that first. On my side the bigger swing came from the runtime and not the model. Could be a workload difference, but I wouldn't generalize too fast from one setup.

[circuitworks90]

Tried to reproduce it just now. I couldn't reproduce the slowdown until I pushed context much higher. The main thing that moved was latency, not quality.

[cacheline]

What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with Kimi and compare from there. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.