[Bug]: multilingual-e5-small performance degradation with multilingual vault #1300
Description
Description
Hi, I have Chinese and English notes in my vault, probably half half. The default model is very fast and works great for English notes, but for Chinese ones it gives irrelevant notes very high scores. I tried switching to multilingual-e5-small to regenerate the embedding and it's fast for a while and at some predicable point slowly grinds to a halt.
The speed before 8000 is somewhere around 3700 token/second, and it's very acceptable, but it keeps slowing down and could get as low as 900. As a result I have never finished the embedding even once because it just takes too long. I'm not entirely sure if this is a bug but I thought to just put it out there.
Steps to Reproduce
No response
Expected Behavior
No response
Actual Behavior
No response
Screenshots
Operating System
No response
Obsidian Version
No response
Smart Connections Version
No response
AI Provider/Model
No response
Other Enabled Plugins
No response
Additional Context
No response