Update llama to pickup "Avoid heavy V transpose operation + improvements"

[deadprogram]

Please see ggml-org/llama.cpp#775

I added issue in this repo since the https://github.com/go-skynet/llama repo doesn't have issues available.

Thanks! 😸

[deadprogram]

Just discovered https://github.com/go-skynet/go-llama.cpp

Closing since all the action is there.

[mudler]

Correct! Seems you already found out all the bits - I will tag soon a new release that should include that as well. Currently images tagged with latest already have the change in, however the API changed quite a bit, mimicking OpenAI api with multi model support (thus docs needs to be updated!)