Thanks (and everyone else) for the reply. I will stick with llama.cpp running llama-server Qwen3-Coder-30B-A3B-Instruct-NVFP4 for now.