I’m using an embedding and cosine similairity system for semantic search purpose.
Since now, i’ve used a model called multi-qa-mpnet-base-dot-v1 from Sentence Transformer package of sbert.net .
You can find the models here, and multi qa is the best at semantic search.
Pretrained Models — Sentence-Transformers documentation
This model is quite small if compared to ada, but it is said that has a performance of 57% in semantic search, that is higher then text-embedding-ada-002 score. But actually these two scores aren’t based on the same database. I doubt that my little model has a performance higher then ada, so i wanted to ask you how can i do to compare their performances since i couldn’t find any semantic search database that is easy to use or if anyone of you knows what of the two is better it would be very helpful for me.
Thanks.