FedGKT client send too large model to the server (MQTT backend)

Hi, I'm trying to test **FedGKT** on raspberry pi 4 using **MQTT** as a communication backend, and I'm confused about some params:
- whether_training_on_client;
- whether_distill_on_the_server;
- sweep.
The issue is that after finishing the first comm-round the clients are unable to send the params to the server because their size is too big ( **more than 4 Gb when trained on client =1** and over 1.4 Gb as a minimum).
How do you explain this and how could resolve this problem?
because even using compression with pickle and Zlib the problem persists.