llama.cpp

c8cdb48d - llama : support all OpenELM models

llama : support all OpenELM models * llama : add variable GQA and variable FFN sizes Some metadata keys can now also be arrays to support setting their value per-layer for models like OpenELM.