Doesn't work for Phi-3 models

#47
by smcleod - opened

For example, trying to create a Q6_K from microsoft/Phi-3-mini-128k-instruct results in:

Error: Error converting to fp16: b'Traceback (most recent call last):\n File "/home/user/app/llama.cpp/convert-hf-to-gguf.py", line 2807, in \n main()\n File "/home/user/app/llama.cpp/convert-hf-to-gguf.py", line 2787, in main\n model_class = Model.from_model_architecture(hparams["architectures"][0])\n File "/home/user/app/llama.cpp/convert-hf-to-gguf.py", line 216, in from_model_architecture\n raise NotImplementedError(f'Architecture {arch!r} not supported!') from None\nNotImplementedError: Architecture 'Phi3ForCausalLM' not supported!\n'

That's odd - let me look into it.

This actually works fine for me: https://huggingface.co/reach-vb/Phi-3-mini-4k-instruct-Q8_0-GGUF
I think the reason for your error was that the official support had not been merged into llama.cpp at that time: https://github.com/ggerganov/llama.cpp/pull/6852

(closing this for now)

reach-vb changed discussion status to closed

Sign up or log in to comment