+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Inference Engines (https://www.datasculpting.ai/inference/forumdisplay.php?fid=9)
+--- Forum: Other Engines (https://www.datasculpting.ai/inference/forumdisplay.php?fid=39)
+--- Thread: prefill bottlenecks on longer prompts (/showthread.php?tid=1432)
Pages:
1
2
prefill bottlenecks on longer prompts - mellowsupply - 10-18-2026
I figured I'd ask here before changing everything.
My current setup is PyTorch with M3 Max, and I keep running into router policy.
Not looking for the perfect answer, mostly what worked in real use.
Re: prefill bottlenecks on longer prompts - quietarch - 10-18-2026
Maybe, but I had the opposite result. I get why that seems right, but on GB200 the bottleneck felt elsewhere.
Re: prefill bottlenecks on longer prompts - deltapoint17 - 10-18-2026
Thanks, that helps. That matches what I saw on 4090.
Re: prefill bottlenecks on longer prompts - tinycontext - 10-18-2026
What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with Gemma and compare from there. Not glamorous advice, but repeatable baselines save a lot of time.
Re: prefill bottlenecks on longer prompts - oxidemixer - 10-18-2026
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.
Re: prefill bottlenecks on longer prompts - amberpatch25 - 10-18-2026
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.
Re: prefill bottlenecks on longer prompts - sanepeak48 - 10-18-2026
Can you fill in a bit more context? What model, what runtime, and roughly what context size?
Re: prefill bottlenecks on longer prompts - kernelghost - 10-18-2026
Did a small comparison this afternoon. My result was mixed: first run looked bad, second run looked normal. I expected the model choice to matter more than it did.
Re: prefill bottlenecks on longer prompts - pixelspark - 10-19-2026
My first guess would be this: Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at INT4. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.
Re: prefill bottlenecks on longer prompts - audiopoint18 - 10-19-2026
Good point. The pattern around batch size has been pretty similar for me. The biggest change for me was simplifying the stack instead of adding one more optimization layer.