@joche_ojeda Structured output is one of those things that looks optional until it isn't. The real gain isn't just clean...

@joche_ojeda Structured output is one of those things that looks optional until it isn't. The real gain isn't just cleaner parsing — it's that you can use smaller, faster local models reliably once you constrain the output shape. JSON schema + llama.cpp grammar sampling works well for this. #LocalAI #LLM