how to create weight dataset in Tnsorflow Federated Learning?

Question

I want to create a weight dataset of the CNN model and using this dataset train an Autoencoder to compress the size of weight updates in a Federated Learning environment, how do get the weights of the model and create a dataset?