Thanks
Title
@athirdpath A new model that claims to reach gpt4 level in terms of writing and roleplay - Starling. I wonder if it is possible to combine noromaid and your lora to it to create something amazing like this guy - Jeonghwan Park did this with his new model : https://huggingface.co/maywell/PiVoT-0.1-Starling-LM-RP
What do you think about this. Will this work or 20b model are better and 7b are not worth trying
And thank you for creating v2. I think the issue of Noromaid writing style not reflected is fixed now and yes this model has better reasoning than hestia!
Yes, that's exactly my next goal! I'm using DPO to get a very similar effect (Their use of "GPT4 Correct User:" implies that's what they did), basically Goofus and Gallant for writing styles. I'm using this model to generate a dataset right now, I should have updates in the next few days.
EDIT: Looked into Starling base, that is exactly what I'm doing, I'm excited to move forward!