Federated predictions

Please help me, I am learning Federated Learning and have this question, do the aggregate server send the global model to each client before the client or worker start training the local model ?
Do the aggregate server make predictions or just aggregation of the local model weights?