Running out of buffer for large models

[MariusGalyan]

I am getting this error message running an 18GB large 30B parameter llm model:
buffer 'data' size 18300780544 is larger than buffer maximum of 17179869184
llama_init_from_file: failed to add buffer

Other llm models with smaller sizes run without any problems. I have a Mac Metal...

[jphme]

I have the same problem and openend an Issue here: ggerganov/llama.cpp#1815

[MariusGalyan]

thank you!