release notes that matter: worth it?

[routecraft]

Spent some time on this today and wanted to compare notes.

I'm comparing two approaches here: keep it simple on Threadripper box, or squeeze more out of the stack and deal with the complexity.

Trying to avoid rebuilding the whole thing twice.

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

[routecraft]

Tried to reproduce it just now. My result was mixed: first run looked bad, second run looked normal. The surprise was that memory pressure showed up before throughput collapsed.

[mangovault67]

What exact setup are you running? Do you see the same thing with a smaller batch or shorter prompt?

[routecraft]

Interesting, I hadn't tried that. I ended up in about the same place using PyTorch. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[modgarden]

I'd approach it in a boring order. If you're on llama.cpp, I'd confirm the version first and then reduce variables until the behavior is obvious. Once you know the plain path works, then add the faster path back in piece by piece. The logs were somehow calmer than I was.

[cacheline]

Interesting, I hadn't tried that. I ended up in about the same place using PyTorch. The biggest change for me was simplifying the stack instead of add

that fixed it for me too.

[routecraft]

Appreciate the detail. That matches what I saw on dual 3090. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[mangopeak72]

Maybe, but I had the opposite result. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit.

[routecraft]

That matches what I was seeing. 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.

[routecraft]

A couple questions before I guess too hard. Are you testing this on A6000 or something else?