Sensei
#1
by aA34543534 - opened
Just an fyi. Model seems to work pretty well with my memory rag. Handles the persona well and handles the web results well in the built in system. (Mornings trigger a website view of NPR up first. Then discussions continue on the daily news and goal setting of the day.) Main thoughts that maybe a MOE system where the RAG part is just one of the experts. Will continue to evaluate the model.
that sounds like an exciting idea, would love to hear if anyone gets a chance to experiment.