FedKD Training Bug

[JY132]

I used FedKD to train on TinyImagenet dataset. Sometimes this bug (LinAlgError: SVD did not converge) will come out from this line: u, sigma, v = np.linalg.svd(param_cpu, full_matrices=False)

[JY132]

The dataset setting is 10 clients, 0.1a, noniid, imbalance,
The training setting is 0.005lr, 64 batch size, 100 training round, 10 local epoch, CNN model