Use simpler prompts for smaller models

[kevin-on]

When using smaller language models, we're seeing inconsistent responses due to the model's limited context handling. To resolve this, we need to optimize and simplify our prompt templates for smaller models.

#45 (comment)

[rom100main]

Hello,
I found that Gemma2 9b SimPO (with nomic-embed-text) managed to get Smart Composer working quite reliably and even with vault chat.
I hope this helps some of you.

[glowingjade]

I'll check it out, thanks!