llama.cpp

0f1bb602 - model : remove duplicate wo_s scale after build_attn (Qwen3, LLaMA) (#22421)