where open models feel surprisingly good

[dockworks23]

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

I've gone back and forth between vLLM and PyTorch for Mixtral. The interesting difference for me has been speculative decoding.

Interested in tradeoffs more than fan loyalty.

[cozysmith]

Good point. The pattern around INT4 has been pretty similar for me.

[northhall88]

Good point. The pattern around INT4 has been pretty similar for me.

What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with FLUX and compare from there.

[routecraft]

Good point. The pattern around INT4 has been pretty similar for me.

A couple questions before I guess too hard. Do you see the same thing with a smaller batch or shorter prompt?

[signaltrack44]

Appreciate the detail. I ended up in about the same place using vLLM. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[northgarden37]

Appreciate the detail. The pattern around consumer GPUs has been pretty similar for me. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[vectorpoint]

A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at INT4.

[kernelarch]

My first guess would be this: On dual 3090, the easiest win is often to back off the aggressive settings and see whether the problem survives.

[flintriver50]

I tested this tonight. My result was mixed: first run looked bad, second run looked normal.

[sanepeak48]

Appreciate the detail. That matches what I saw on 4090. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.