tokens/sec tuning ideas

[routeline23]

Posting this in case it saves someone else a little time.

I'm not trying to build a perfect architecture diagram. I just want a sane starting point around MLX and B200.

I'm especially interested in mistakes worth avoiding.

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

[routecraft]

My first guess would be this: 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.

[sparrowkite53]

That matches what I was seeing. I ended up in about the same place using vLLM.

[cobaltspark]

I'd start with the simpler stack.

I'm happy to test one more variant if you want a second data point.

[routecraft]

Appreciate the detail. I don't think your conclusion is far off. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[sanescope]

What worked for me was pretty simple. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at speculative decoding. Not glamorous advice, but repeatable baselines save a lot of time.