draft model choices for Phi

[cacheline]

Thought I'd post a small recipe because a lot of threads ask similar things.

This isn't revolutionary, but the combination of LM Studio and 5090 felt surprisingly usable once I tuned around quant drift.

If someone wants the exact settings I can clean them up and post them.

[fastfield4]

Thought I'd post a small recipe because a lot of threads ask similar things.

Tried to reproduce it just now. On 5090 with SGLang, I saw basically the same behavior. I expected the model choice to matter more than it did.

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

[cacheline]

Good point. 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.

[vramarch]

That matches what I was seeing. I don't think your conclusion is far off. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[cacheline]

A couple questions before I guess too hard. Is this local-only or are you routing anything externally?

[cobaltworks25]

subscribed, curious where this lands.

[cacheline]

Can you fill in a bit more context? Do you see the same thing with a smaller batch or shorter prompt?

[promptpoint]

My first guess would be this: Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at KV cache pressure. Once you know the plain path works, then add the faster path back in piece by piece.

[amberrunner]

Ran a quick check on my box. My result was mixed: first run looked bad, second run looked normal. The main thing that moved was latency, not quality.

[cacheline]

A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at HBM constraints. Not glamorous advice, but repeatable baselines save a lot of time.