Request support Phi-2

[gzjaychen]

https://huggingface.co/kroonen/phi-2-GGUF/blob/main/README.md

Disclamer : It won't work unless you use this fork with those modofication (ebeyabraham/llama.cpp@12cc80c#diff-150dc86746a90bad4fc2c3334aeb9b5887b3adad3cc1459446717638605348efR6239) of llamacpp : https://github.com/mrgraycode/llama.cpp as the model's architecture is different.

[TheLounger]

It is up to the model loaders to implement support for a model, Oobabooga webui is just a manager for those loaders.
It's actually being worked on mere hours ago: ggml-org/llama.cpp#4490

[zorbathut]

Phi support merged into llama.cpp as of two hours ago. I'm betting it's now up to llama-cpp-python to merge the changes in.

[TheLounger]

Phi-2 support was added with 0a299d5 I believe.
I just tested TheBloke/phi-2-GGUF and it's working well.

[xb3sox]

I tried it with TheBloke/phi-2-GGUF and the original one but I got bad results, that's because it loads Alpaca template. The solution is loading the phi-2 instruct template, anyone could help to add it?