Evaluating LLM-generated code for domain-specific languages

Any experience out there wrt LLMs generating DSL code (your story or hard data both welcome)?