FedMABench is an open-source benchmark for federated training and evaluation of mobile agents, specifically designed for heterogeneous scenarios.
FedMABench includes the following key features:
6 datasets with 30+ subsets （over 800 apps across 5 categories）
3 types of heterogeneity (e.g., App Category Distribution, Specific App… See the full description on the dataset page:
https://huggingface.co/datasets/wwh0411/FedMABench.
