+- 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: embedding model you trust (/showthread.php?tid=186)
embedding model you trust - localtinker - 09-09-2026
Did a bit of testing and thought I'd post the result here.
I tried this with Phi using SGLang on M3 Max. The headline result was better than I expected, but structured output still ended up being the limiting factor.
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.
Re: embedding model you trust - cudaline - 09-09-2026
Maybe, but I had the opposite result. I wouldn't optimize around that first. On my side the bigger swing came from the runtime and not the model.
Re: embedding model you trust - kernelmike - 09-09-2026
A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with Kimi and compare from there. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.
Re: embedding model you trust - pixelcrate82 - 09-12-2026
Ran a quick check on my box. My result was mixed: first run looked bad, second run looked normal. I expected the model choice to matter more than it did.
Re: embedding model you trust - localpatch78 - 09-13-2026
Quote:Ran a quick check on my box. My result was mixed: first run looked bad, second run looked normal. I expected the model choice to matter more than it d
That matches what I was seeing. That matches what I saw on M3 Max. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing. I keep promising myself this is the last rebuild.
Re: embedding model you trust - sparrowgrid - 09-16-2026
Maybe, but I had the opposite result. I've seen people blame tail latency when the real issue was much more mundane. Worth testing the simpler explanation before adopting the more interesting one.