llama.cpp

llama : fix loading models with shared tok_embd and output

#5651

Merged