on-device inference worth the effort: what are you seeing?

[amberrunner]

This keeps coming up in conversation so I figured I'd start a thread.

One thing I keep noticing is that people underestimate consumer GPUs until they actually try to run this stuff for more than one afternoon.

Maybe obvious to some of you, but it clicked for me this week.

This is mostly for local work rather than public serving.

The part that surprised me was how much the boring details mattered more than the model choice.

[cudacrate]

This keeps coming up in conversation so I figured I'd start a thread.

Appreciate the detail. I ended up in about the same place using PyTorch. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[plainline]

My first guess would be this: If you're on vLLM, I'd confirm the version first and then reduce variables until the behavior is obvious. Once you know the plain path works, then add the faster path back in piece by piece.

[sparrowshed]

I don't think that's the whole story. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit.

[orbitbench6]

That matches what I was seeing. That matches what I saw on GB200. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.

[sparrowshed]

Ran a quick check on my box. The issue went away once I changed one setting and restarted the stack cleanly.

[plainline]

Thanks, that helps. 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.

[sparrowghost]

What exact setup are you running? Is this local-only or are you routing anything externally?

There are a few different conversations hiding inside this thread, so being explicit about the goal helps.

[llmpeak]

worth checking the driver first.

[marbleyard32]

Tried to reproduce it just now. The issue went away once I changed one setting and restarted the stack cleanly. The main thing that moved was latency, not quality.