when a plain Python stack is enough for Qwen

[fastbench]

That matches what I was seeing. The pattern around MoE serving has been pretty similar for me. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[kernelpeak]

A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with Mistral and compare from there. Turns out 'just one more test' was a lie.

[homelabhenry]

Good point. I don't think your conclusion is far off. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable. At this point the real bottleneck is my patience.

[audioforge]

Interesting, I hadn't tried that. I don't think your conclusion is far off. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

If you post your exact settings, people here can probably narrow it down pretty fast.

[turbogrid11]

Interesting, I hadn't tried that. I don't think your conclusion is far off. The biggest change for me was simplifying the stack instead of adding one

A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with Phi and compare from there.