Unknown model architecture: 'phi2'

[SalihErenYzb]

It currently does not work with Microsoft's model phi-2. Saw a similar issue on llama.cpp. It looks like they fixed it, is there a way to use the new llama.cpp with llamafile? Sry If it is stupid I'm a newbie

[SalihErenYzb]

I can just build from source...🫠

[jkeys089]

The relevant llama.cpp change can be found at: ggml-org/llama.cpp#4490

@SalihErenYzb I'm not 100% clear how to pull in upstream changes from llama.cpp. Would you mind sharing the process you used to get this working?

@jart Do you have any plans to stay (relatively) up to date with upstream llama.cpp changes?

[aamott]

I second that it would be great to have phi 2 supported. It's pretty awesome for being so small, and might be worth putting on the front page.