mobile NPU reality check

[marbleyard32]

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

I've gone back and forth between TensorRT-LLM and llama.cpp for Mistral. The interesting difference for me has been quant drift.

Interested in tradeoffs more than fan loyalty.

[cudacrate]

I'd start with the simpler stack.

[torchkite]

I'd start with the simpler stack.

What worked for me was pretty simple. If you're on TensorRT-LLM, 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.

[cozysmith]

I'd start with the simpler stack.

What worked for me was pretty simple. On 7900 XTX, the easiest win is often to back off the aggressive settings and see whether the problem survives.

[relaysupply90]

Good point. 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.

[promptspark58]

Ran a quick check on my box. I couldn't reproduce the slowdown until I pushed context much higher. The surprise was that memory pressure showed up before throughput collapsed.

[llmpeak]

Not sure I agree there. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit.