paged attention observations

[cacheline]

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

It feels like the conversation has shifted hard toward inference and serving details lately, and honestly that makes sense once you start dealing with workstation thermals.

Maybe obvious to some of you, but it clicked for me this week.

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

[quantyard]

that fixed it for me too.

[cacheline]

that fixed it for me too.

Thanks, that helps. The pattern around NVFP4 has been pretty similar for me. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[vectorcrate98]

that fixed it for me too.

What worked for me was pretty simple. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at prefix caching.

[sparrowghost]

A couple things I'd check first. 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.

[cacheline]

Tried to reproduce it just now. I couldn't reproduce the slowdown until I pushed context much higher. The surprise was that memory pressure showed up before throughput collapsed.

[fusedriver]

Tried to reproduce it just now. I couldn't reproduce the slowdown until I pushed context much higher. The surprise was that memory pressure showed up

Did a small comparison this afternoon. My result was mixed: first run looked bad, second run looked normal.

[cacheline]

A couple questions before I guess too hard. What model, what runtime, and roughly what context size?