Open nonetrix opened 2 weeks ago
Try turning off the 'Chat with website using vector embeddings' setting for summarizing; it will work much better than RAG-based methods. Also, I think you may have turned it off in Firefox, which is why it's giving higher-quality response
It was enabled in Firefox too oddly, doesn't seem to have much of a effect really. I am not sure how RAG exactly works, but can I get raw text that's passed to the model perhaps to debug? Also, seemingly I haven't had issue with it really
Not sure why but RAG is much worse in Firefox than Brave
Firefox:
Brave:
I made sure I had the same model and settings on both, not sure what is causing the difference. Perhaps some privacy setting Brave has or Manifest v3