ludwig-ai/ludwig ? reverse-engineered prompt

Build me a Python library and command line tool that lets people train, fine tune, predict with, and serve AI models from a simple YAML config instead of writing a lot of code.

It should support common use cases like text models, tabular prediction, and multimodal models with images plus text. I want to be able to point it at a CSV or dataset, define inputs and outputs in a config file, run a train command, then use the saved model for predictions or a small local API server. It would also be great if it can generate a starter config from a plain English description so someone can get going fast.

Please make it feel polished, with good defaults, clear error messages, examples, and tests. Use current best practices for modern Python ML, and look up the latest docs online if you need to.