Pysyft integration with pytorch

1

I am using the pysyft library 0.8.2 to implement Federated Learning experiment. At one part of the code I am supposed to send the tensor information over clients established using pysyft commands. But the attributes send() and get() are no longer supported by Pytorch 2.1.0.

import torch
import pysyft as sy

bob = sy.Worker(name=‘bob’)
print(bob)
print(bob.client_cache)
x = torch.tensor([1,2,3])
print(x)
x.send(bob) # error message: ‘Tensor’ objects has no attributes ‘send’
print(x)

How to overcome this dilemma and send tensor over the clients?

2

I don’t think send and get were ever supported tensor functions so I assume these were defined in custom tensor classes?

3

As the following link indicates: Link using hook = sy.TorchHook(torch) all torch tensors will get additional functionality - methods like .send() and get() . The issue now is that in new Pysyft library sy.TorchHook(troch) is no longer needed, therefore, how to extend the torch functionality to include send() and get(). Or what is the replacement for such commands to perform the same action?