good hosted AI app that still respects privacy for Mixtral

[mangopeak72]

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

Just sharing a setup that ended up working reasonably well: Llama, PyTorch, M3 Max, with a lot of attention on continuous batching.

Maybe useful as a starting point for newer folks.

[mangopath4]

Maybe, but I had the opposite result. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit. Could be a workload difference, but I wouldn't generalize too fast from one setup.

[torchsmith25]

A couple questions before I guess too hard. Are you testing this on 5090 or something else?

[mangopeak72]

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

[audiodesk]

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

My first guess would be this: Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at NVFP4. Once you know the plain path works, then add the faster path back in piece by piece.

[rocmpath]

Ran a quick check on my box. My result was mixed: first run looked bad, second run looked normal. I expected the model choice to matter more than it did.

[mangopath4]

Ran a quick check on my box. My result was mixed: first run looked bad, second run looked normal. I expected the model choice to matter more than it d

Interesting, I hadn't tried that. That matches what I saw on Threadripper box. The biggest change for me was simplifying the stack instead of adding one more optimization layer.