train.optimizer.lr: type: cate choices: [ 1e-1, 1e-2, 1e-3, 1e-4 ] train.local_update_steps: type: cate choices: [1,2,3,4]