Phi-3-mini support

[kangurek-kao]

Hello!
First of all, I'm impressed by this project and I hope it will pick up some steam in the nearby future. But I would really like to see additional LLMs/SLMs, as local models get faster and more creative every single day. Like Microsoft's Phi-3 models.

https://news.microsoft.com/source/features/ai/the-phi-3-small-language-models-with-big-potential/

Apparently they are open-sourced and they are comparable performance-wise to vanilla GPT-3 and Llama models. They can be run locally though, meaning they fit perfectly into this project (especially when Phi-1.5 is already implemented).

I would love seeing it, although I'm afraid I couldn't help too much outside of translating things. Thus consider this issue as an idea rather than issue per se.

Warm greetings

[abi]

I'm planning on supporting phi3 soon. Maybe, this weekend.