The metrics are slowly leaving Datasets (they are being deprecated as we speak) to move to the Evaluate library. We are looking for contributors to help us with the move.

Normally, the migration should be as easy as replacing the import of load_metric from Datasets to the same function in Evaluate. To fix all tests, a dependency to evaluate will need to be added in the requirements file (this is the link for PyTorch, there is another one for the Flax examples).

If you're interested in contributing, please reply to this issue with the examples you plan to move.