prefill bottlenecks on longer prompts: what are you seeing?

[oxidemixer]

Did a bit of testing and thought I'd post the result here.

After a few runs on A6000, I ended up with decent performance, but the setup only really made sense once I adjusted around workstation thermals.

Nothing scientific here, just practical notes from a real setup.

My use case is pretty ordinary: local chat, some document work, and occasional batch jobs.

[vrampilot]

What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with DeepSeek and compare from there. Once you know the plain path works, then add the faster path back in piece by piece.

[phaseshed]

I tested this tonight. The issue went away once I changed one setting and restarted the stack cleanly.

[cobalttinker]

My first guess would be this: On 5090, the easiest win is often to back off the aggressive settings and see whether the problem survives. Not glamorous advice, but repeatable baselines save a lot of time.

[batchfield55]

Did a small comparison this afternoon. My result was mixed: first run looked bad, second run looked normal. The main thing that moved was latency, not quality.

[southspark]

Did a small comparison this afternoon. My result was mixed: first run looked bad, second run looked normal. The main thing that moved was latency, not

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

[batchsize]

[quote]Did a small comparison this afternoon. My result was mixed: first run looked bad, second run looked normal. The main thing that moved was laten

Ran a quick check on my box. On mini PC with PyTorch, I saw basically the same behavior. The main thing that moved was latency, not quality.

If you post your exact settings, people here can probably narrow it down pretty fast.

[torchkite]

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. Not glamorous advice, but repeatable baselines save a lot of time.

[mangoquill]

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

Need a little more detail here. Are you testing this on B200 or something else?

[plainline]

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