Your documentation is super well-written, so it was really easy to convert the framework. My only request is that I would like some sort of tokenizer so I can be more precise with the amount of tokens for input and output and build a more robust system around that 🙂