llama.cpp

llama : make Qwen2MoE QKV bias optional

#12477

Merged