llama.cpp

b457f89e - server: add --reasoning-format=disabled to disable thinking (incl. qwen3 w/ enable_thinking:false)