[Enhancement]: Add seperate prompts folders for local llm's based on token limits.

[Wladastic]

Version

VisualStudio Code extension

Suggestion

Currently all the prompts are only for GPT-4-Turbo, but using them with local llm's is currently impossible.

I would suggest adding a whole new folder structure that allows less tokens to be used or even one per architecture: GPT, LLama, Mistral, ( I know its llama too but works different) Gemma, etc.

For each of those we can have then curated prompts or even a subfolder: <model_arc_folder>/

• 1k/ (very experimental but would allow 1B or 3B models to work)
• 2k/ (especially gemma 2b)
• 4k/ (llama or GPT-3.5-Turbo)
• 8k/ (mistral 7B instruct 0.1 based or GPT-4)
• 16k/ (Mixtral 4x7B or 8x7B)
• 32k/ (mistral 7B instruct 0.2 based or Mixtral)
• 64k/ (yarn mistral 7B or 13B models)
• 128k/ (GPT-4 Turbo level)

This way its easier to also add benchmarking to each level of those llm prompts and models.