when a plain Python stack is enough for FLUX

[nanogarden]

that fixed it for me too. The logs were somehow calmer than I was.

Thanks, that helps. That matches what I saw on M3 Max.

[southpatch]

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

[cobalttinker]

Ran a quick check on my box. On 7900 XTX with llama.cpp, I saw basically the same behavior. The main thing that moved was latency, not quality.

[promptpoint]

Not sure I agree there. I've seen people blame consumer GPUs when the real issue was much more mundane. Worth testing the simpler explanation before adopting the more interesting one.

[cacheline]

I think the bottleneck is somewhere else. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit.

Can you fill in a bit more context? Are you testing this on dual 3090 or something else?

[vramridge]

Appreciate the detail. I don't think your conclusion is far off.