Question about FedMKT example

[Sahar-ghoflsaz]

Hi, I have a question about the FedMKT implementation.

I am trying to understand exactly what information is sent from each client to the server during training.

From my reading, it seems that clients should send some form of public logits/predictions computed on the public dataset, which are then aggregated or used by the server for knowledge transfer. However, when I inspect the public logits in my run, they appear to be either NaN or blank / whitespace.

So I would like to clarify:

1 - What is the exact object that each client sends to the server in the current FedMKT implementation? How are they organized?

2 - At what step in the code are these values produced and transmitted? Are they all transmitted at the end of the local training?

3 - Under what conditions could the public logits become NaN or empty?

For context, I am debugging the implementation because I want to verify the client-to-server communication flow, and right now, the public logits I inspect do not look valid.

Thank you.