@yixuantt might be interested in adding the integration?

Thanks for the kind words! Unfortunately, I may not have enough bandwidth to finish it recently. (Maybe later if needed)

Here is a short description of the difference between these two benchmarks: almost all task-evaluating processes are similar except for summarization. The summarization is based on the text and summary instead of the "human summary vs the machine summary". For other tasks, simply copying the task dataset metadata is fine.

Thanks, @yixuantt great to have those details!