allenai/allennlp ? reverse-engineered prompt

Build me an open source Python library for NLP research on top of PyTorch, like a clean framework that makes it easy to define models, train them, evaluate them, and reuse components across different language tasks.

I want it to feel ready for real experiments, with a simple command line workflow, config driven training, support for custom modules and plugins, and good docs so someone can get started without guessing. Include a solid test suite, examples or templates for typical NLP projects, and the kind of utilities researchers expect for data handling, metrics, and model saving and loading.

Please keep it practical and polished, with clear documentation and a structure that makes it easy to extend. If you need to check anything recent, look up current docs online first.