Finding system prompts and special tokens for open LLMs

Models
1

I want to try some of the top models from the Open LLM Leaderboard for a series of tasks that require the models to reply to prompts. The models are:

  • MetaIX/GPT4-X-Alpasta-30b
  • digitous/Alpacino30b
  • Aeala/GPT4-x-AlpacaDente2-30b
  • TheBloke/dromedary-65b-lora-HF

According to some folks who responded to a github issue, being able to pass a system prompt and also to add special tokens to the start and end of requests is paramount to getting reasonable model performance.

However, I’m unable to figure out from the model cards and model documentation for these models; the only example I’ve found is for Open-Assistant, which doesn’t rank among the top LLMs.

Wondering if anyone knows where I can find this kind of information?

EDIT: formatting