Request for Code Sharing and Client Dataset Information

[geehokim]

I’ve been trying to reimplement the method, but I'm facing some challenges. The GCL loss is not training properly, and with the current Dirichlet implementation, it's impossible to obtain the reported distribution from the table. For instance, the Dirichlet setting with α = 0.2 is quite heterogeneous, and it's difficult for a single client to sample all classes as unlabelled data.

Could you please share the code or the client dataset information used in the original implementation? It would help me a lot in understanding and troubleshooting the issue.

Thank you!