Reducing Number of Prompt Cycles to improve efficiency while not impacting response quality

[Govind-S-B]

Right now at certain parts , the input is generated via 2 steps . One is the content generation and the other is a formatter which are two steps to just generate a list of items to be rendered on the presentation.

Trying new models and implementing the Grammar functionality from llama.cpp might help with this situation of reducing the number of prompts needed

[Govind-S-B]

Note : OpenHermes 2.5 mistral is really good at what you specify , need to restructure current prompts entirely if working with openhermes