In Fed-ML HE example, the client model weights are not encrypted.
mayank64ce opened this issue · comments
The fedml-he example as implemented here does not seem to encrypt the model weights (torch tensors are communicated instead of ciphertext) even when the enable_fhe
flag is true
in the config file.
I checked the source code here and found that on line 29, the line should be:
if self.is_enabled:
return
Is that a mistake or did I interpret the code wrong ?