This is a Python library that constrains the output format of large language models, ensuring results conform to the specified structure during generation, rather than patching with regular expressions after generation. When called, you only need to pass in the expected type, and the same code can seamlessly switch among OpenAI, Ollama, and vLLM