Audio / Speech discussion

[vramport]

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

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

Curious whether that lines up with what other people are seeing.

[sparrowhall]

Can you fill in a bit more context? What model, what runtime, and roughly what context size?

[macrostack69]

Not sure I agree there. I've seen people blame prefill/decode split when the real issue was much more mundane. Not saying you're wrong, just that I wouldn't lock that in as the lesson yet.

[vectorfield]

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.

[signalcraft14]

Appreciate the detail. I ended up in about the same place using TensorRT-LLM. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[bytesupply76]

same here on my side.

[vectorfield]

I'd approach it in a boring order. On A6000, the easiest win is often to back off the aggressive settings and see whether the problem survives. Once you know the plain path works, then add the faster path back in piece by piece.

[cozywire]

Did a small comparison this afternoon. The issue went away once I changed one setting and restarted the stack cleanly.