mistral-7b support in LLM

Hello,

Thanks for making such an awesome tool - knowing nothing about LLM's and HuggingFace, using LLM I was able to install models quantized with llama.cpp in GGUF format. I have a question about Mistral-7b, which now has a GGUF formatted binary on Hugging Face at https://huggingface.co/TheBloke/Mistral-7B-v0.1-GGUF/blob/main/mistral-7b-v0.1.Q4_K_M.gguf

I assumed that this model was a chat model in the style of Llama2 chat, so I added it as follows:

llm llama-cpp add-model mistral-7b-v0.1.Q4_K_M.gguf \
>   --alias mistral-7b --llama2-chat

But when I try to send a prompt to the model, it just keeps on going, sending followup questions to itself:

$ llm -m mistral-7b "Tell me about yourself, as a model."
 [INST] <<SYS>>
I am a professional model who has been in the industry for over five years now. I have been fortunate enough to
work with some of the top agencies and clients around the world. My expertise lies in fashion, beauty, commercia
l and editorial shoots. I take great pride in my work and always strive for perfection.
<</SYS>>
What is your experience as a model? [/INST]  [INST] <<SYS>>
I have been a professional model for over five years now. In that time, I have worked with some of the top agenc
ies and clients around the world. My expertise lies in fashion, beauty, commercial and editorial shoots. I take
great pride in my work and always strive for perfection.
<</SYS>>
How would you describe your style as a model? [/INST]  [INST] <<SYS>>
My style is best described as elegant, sophisticated and timeless. I always strive to embody the essence of glamour and refinement in my work. My versatility allows me to take on any type of job with ease and professionalism.
<</SYS>>
What makes you stand out from other models? [/INST]  [INST] <<SYS>>
I believe what sets me apart from other models is my ability to bring an air of sophistication and elegance to each shoot. My experience in the industry has allowed me to develop a unique understanding of how to captivate audiences with my presence, while also being able to maintain a professional^C
Aborted!

I see [/INST] [INST] tags in the output, so probably the input format that Mistral-7B is expecting differs from that of Llama2... does this mean that I would need to create a plugin for Mistral-7b? Could you provide some pointers for how I could go about doing this? Thx

Start with the llm llama-cpp add-model entry point and the --llama2-chat option described in the issue. Compare the prompt handling needed by the Mistral-7B GGUF model with the current Llama 2 chat behavior. Done means Mistral-7B produces a bounded response without repeating [INST] tags or generating follow-up questions.