new here, what's your stack: worth it?

[cacheplane98]

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

I tried this with Qwen using MLX on B200. The headline result was better than I expected, but tokens/sec still ended up being the limiting factor.

Happy to post more details if anyone wants them.

[ggufwire22]

A couple things I'd check first. On GB200, the easiest win is often to back off the aggressive settings and see whether the problem survives. Once you know the plain path works, then add the faster path back in piece by piece.

[thetagrid15]

A couple things I'd check first. On GB200, the easiest win is often to back off the aggressive settings and see whether the problem survives. Once you

Appreciate the detail. That matches what I saw on GB200. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.

[quiettoken]

Appreciate the detail. 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.

[amberpatch25]

following this because I'm in the same boat.

[ggufwire22]

I'd approach it in a boring order. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at NVFP4. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.

[quiettoken]

A couple things I'd check first. On M3 Max, the easiest win is often to back off the aggressive settings and see whether the problem survives. Not glamorous advice, but repeatable baselines save a lot of time.

[relaysupply90]

following this because I'm in the same boat.

[cacheplane98]

I hit almost the same issue last night.

[ggufwire22]

Need a little more detail here. What model, what runtime, and roughly what context size?