question about how rpc works in the hivemind package

[drimeF0]

Hello, I am currently using hivemind for my project to distribute MoE models with huggingface transformers to peer-to-peer network participants for their inference and training. At the moment I'm stuck because I don't quite understand how rpc calls work in the hivemind package. I wanted to implement the ability to download the state_dict of each expert from network participants, but I still didn’t quite understand how to do it correctly. I assume that in "moe.server.connection_handler.ConnectionHandler" you need to create a method that starts with rpc_, and then use stub in "moe.client.expert"?