llama.cpp

model : add QKV weight fusion for LLaMA, Qwen2, and Qwen3

#20628

Open