tokens/sec tuning ideas for Kimi

[batchsize]

This keeps coming up in conversation so I figured I'd start a thread.

One thing I keep noticing is that people underestimate tool calling until they actually try to run this stuff for more than one afternoon.

Curious whether that lines up with what other people are seeing.

[sparrowshed]

What worked for me was pretty simple. If you're on MLX, I'd confirm the version first and then reduce variables until the behavior is obvious.

[amberpatch25]

I'd approach it in a boring order. If you're on PyTorch, I'd confirm the version first and then reduce variables until the behavior is obvious. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.

[edgecoast]

Appreciate the detail. That matches what I saw on B200. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

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

[quietcraft]

I'd approach it in a boring order. Before rebuilding anything, I'd test one clean baseline with Qwen and compare from there. Once you know the plain path works, then add the faster path back in piece by piece.

[vramport]

Can you fill in a bit more context? Is this local-only or are you routing anything externally?

[cudapeak]

I tested this tonight. I couldn't reproduce the slowdown until I pushed context much higher.

[vramforge]

following this because I'm in the same boat.