human evals on small teams for Phi

[vectorcrate98]

I tested this tonight. The issue went away once I changed one setting and restarted the stack cleanly.

[kernelarch]

A couple things I'd check first. On mini PC, the easiest win is often to back off the aggressive settings and see whether the problem survives.

[oxideworks59]

worth checking the driver first.

[promptpoint]

that fixed it for me too.

[oxideworks59]

Can you fill in a bit more context? Are you testing this on 7900 XTX or something else?

A couple things I'd check first. 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.

[vrambandit]

subscribed, curious where this lands. I keep promising myself this is the last rebuild.

[rocmriley]

Not sure I agree there. I've seen people blame speculative decoding when the real issue was much more mundane. Not saying you're wrong, just that I wouldn't lock that in as the lesson yet.

[kernelgrid]

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

The more I work with local setups, the more it seems like continuous batching decides everything downstream.

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

[promptpoint]

Appreciate the detail. That matches what I saw on 5090. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[bytepilot]

Good point. I ended up in about the same place using LM Studio. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.