Hey -- I'm enjoying using llama-index very much for my project, and have it successfully deployed backing a chatbot. Now, I'm hoping to experiment and improve the model's responses to various questions, and if possible reduce the cost, as it's about 7c a query at the moment, which is too expensive for me to run publicly. Does anyone have experience with iterating through different parameterizations to improve model performance or reduce cost? I'm working with a substantial custom corpus (25MB) of somewhat high complexity.
