batching changes after update for Llama

[kernelmike]

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

I've gone back and forth between MLX and llama.cpp for Kimi. The interesting difference for me has been prefill/decode split.

What would you pick if you had to live with it for a few months?

My use case is pretty ordinary: local chat, some document work, and occasional batch jobs.

[kernelmike]

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.

[thetapath]

Interesting, I hadn't tried that. The pattern around quant drift has been pretty similar for me.

[audioriver]

Interesting, I hadn't tried that. The pattern around quant drift has been pretty similar for me.

I'd approach it in a boring order. Before rebuilding anything, I'd test one clean baseline with Qwen and compare from there. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.

[thetapath]

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

[amberpatch25]

A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with Mixtral and compare from there. Not glamorous advice, but repeatable baselines save a lot of time.

[edgecoast]

That matches what I was seeing. That matches what I saw on M3 Max. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[kernelmike]

That matches what I was seeing. That matches what I saw on M3 Max. Worth rechecking your defaults though, because one bad assumption can make the rest

Need a little more detail here. Is this local-only or are you routing anything externally?

[edgecoast]

My first guess would be this: On Threadripper box, the easiest win is often to back off the aggressive settings and see whether the problem survives. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.

[dockriver]

My first guess would be this: On Threadripper box, the easiest win is often to back off the aggressive settings and see whether the problem survives.

That matches what I was seeing. 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.