Thanks! The model is trained only for O0-3, not support for obfuscation. There's still a long way for llm to go.