AI2 Tango allows you to build machine learning experiments out of steps that can be reused and repeated. Users can write their own steps simply by wrapping common Python functions. Tango also comes with a library of pre-built steps for training models, working with datasets, and running evaluations. It is integrated with popular tools like the Huggingface transformers library, PyTorch Lightning, and others.
Tango’s built-in mechanism for storing and retrieving results makes sure that researchers can stay flexible when pursuing another idea. No work is duplicated, past results can be found easily, and the way a result was obtained is stored along with the result itself.