+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Generative AI (https://www.datasculpting.ai/inference/forumdisplay.php?fid=17)
+--- Forum: Audio / Speech (https://www.datasculpting.ai/inference/forumdisplay.php?fid=71)
+--- Thread: Help with Audio / Speech (/showthread.php?tid=821)
Help with Audio / Speech - signalcraft14 - 10-06-2026
Posting this in case it saves someone else a little time.
I'm not trying to build a perfect architecture diagram. I just want a sane starting point around SGLang and Threadripper box.
Would rather learn from your scars than collect my own all at once.
Re: Help with Audio / Speech - sparrowkite53 - 10-08-2026
I tested this tonight. My result was mixed: first run looked bad, second run looked normal.
Re: Help with Audio / Speech - vramforge - 10-08-2026
Ran a quick check on my box. On 5090 with llama.cpp, I saw basically the same behavior. The main thing that moved was latency, not quality.
Re: Help with Audio / Speech - batchpeak - 10-08-2026
A couple questions before I guess too hard. Are you testing this on mini PC or something else?
Re: Help with Audio / Speech - cudapoint - 10-08-2026
Quote:A couple questions before I guess too hard. Are you testing this on mini PC or something else?
Maybe, but I had the opposite result. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit. Not saying you're wrong, just that I wouldn't lock that in as the lesson yet.
Re: Help with Audio / Speech - flintriver50 - 10-10-2026
What worked for me was pretty simple. On A6000, the easiest win is often to back off the aggressive settings and see whether the problem survives.
Re: Help with Audio / Speech - vramport - 10-10-2026
worth checking the driver first.
Re: Help with Audio / Speech - vramridge - 10-11-2026
I'd start with the simpler stack.
Re: Help with Audio / Speech - sparrowhall - 10-12-2026
Appreciate the detail. The pattern around prefix caching has been pretty similar for me.
Re: Help with Audio / Speech - cozywire - 10-12-2026
A couple questions before I guess too hard. Do you see the same thing with a smaller batch or shorter prompt?