Replace a specific attribute in RecursiveScriptModule

jit
1 
import torch
from torch import nn

class ADD1(nn.Module):
    def __init__(self):
        super().__init__()
    def forward(self, x):
        return x + 1

class ADD2(nn.Module):
    def __init__(self):
        super().__init__()
    def forward(self, x):
        return x + 2

class TEST(nn.Module):
    def __init__(self):
        super().__init__()
        self.add = ADD1()
    def forward(self, x):
        return self.add(x)

x = torch.zeros(1)

test = torch.jit.script(TEST().eval())
print(test.graph)
print(test(x))
# tensor([1.])

new_add = torch.jit.script(ADD2().eval())

# test.add = new_add

# Traceback (most recent call last):
#   File "test2.py", line 41, in <module>
#     test.add = new_add
#   File "/usr/local/lib/python3.8/site-packages/torch/jit/_script.py", line 791, in __setattr__
#     self._modules[attr] = value
#   File "/usr/local/lib/python3.8/site-packages/torch/jit/_script.py", line 244, in __setitem__
#     self._c.setattr(k, v)
# RuntimeError: Expected a value of type '__torch__.ADD1 (of Python compilation unit at: 0x55b702ec2cb0)' for field 'add', but found '__torch__.ADD2 (of Python compilation unit at: 0x55b702ec2cb0)'

test.add._reconstruct(new_add._c)
print(test.graph)
print(test(x))
# tensor([1.])

How to replace test.add with new_add in torchscript? I tried test.add = new_add and test.add._reconstruct(new_add._c) , but neither of them works.
Thanks