+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Training & Customization (https://www.datasculpting.ai/inference/forumdisplay.php?fid=13)
+--- Forum: Dataset Creation / Curation (https://www.datasculpting.ai/inference/forumdisplay.php?fid=56)
+--- Thread: reranking still worth it for small sets: early impressions (/showthread.php?tid=1376)
reranking still worth it for small sets: early impressions - fastlab - 10-17-2026
Posting this in case it saves someone else a little time.
I'm helping a small team set up a local workflow. We don't need a giant stack, but we do care about MoE serving.
Would rather learn from your scars than collect my own all at once.
Re: reranking still worth it for small sets: early impressions - vectorshed43 - 10-18-2026
I'd approach it in a boring order. On 7900 XTX, 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. I keep promising myself this is the last rebuild.
Re: reranking still worth it for small sets: early impressions - quiettoken - 10-18-2026
A couple questions before I guess too hard. What model, what runtime, and roughly what context size?
Re: reranking still worth it for small sets: early impressions - rocmriley - 10-18-2026
Interesting, I hadn't tried that. 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.
Re: reranking still worth it for small sets: early impressions - sparrowhall - 10-18-2026
Good point. I don't think your conclusion is far off.
Re: reranking still worth it for small sets: early impressions - homelabhenry - 10-18-2026
that fixed it for me too.
Re: reranking still worth it for small sets: early impressions - audioriver - 10-18-2026
A couple questions before I guess too hard. Are you testing this on B200 or something else?
Re: reranking still worth it for small sets: early impressions - llmtrack26 - 10-19-2026
Ran a quick check on my box. On GB200 with SGLang, I saw basically the same behavior. The main thing that moved was latency, not quality.