incompatible with PyTorch 1.9

I understand I should use a lower grade version of PyTorch. I just want to use CrypTen to compute the inner product. Can anyone help me to fix it with a few efforts? Really appreciate it! Thanks!

My env: torch 1.9.0, torchvision 0.10.0

Here is the code

import torch
import crypten

crypten.init()
x = torch.tensor([1.0, 2.0, 3.0])
x_enc = crypten.cryptensor(x) # encrypt
z_enc = x_enc.dot(x_enc)

And the error

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
/tmp/ipykernel_31576/2452821068.py in <module>
      5 x = torch.tensor([1.0, 2.0, 3.0])
      6 x_enc = crypten.cryptensor(x) # encrypt
----> 7 z_enc = x_enc.dot(x_enc)
      8 print(z_enc.get_plain_text())

~/.conda/envs/py38/lib/python3.8/site-packages/crypten/mpc/mpc.py in convert_wrapper(self, *args, **kwargs)
     40             def convert_wrapper(self, *args, **kwargs):
     41                 result = self.to(ptype)
---> 42                 return func(result, *args, **kwargs)
     43 
     44             return convert_wrapper

~/.conda/envs/py38/lib/python3.8/site-packages/crypten/mpc/mpc.py in ob_wrapper_function(self, value, *args, **kwargs)
   1141         if isinstance(value, CrypTensor):
   1142             value = value._tensor
-> 1143         result._tensor = getattr(result._tensor, name)(value, *args, **kwargs)
   1144         return result
   1145 

~/.conda/envs/py38/lib/python3.8/site-packages/crypten/mpc/primitives/arithmetic.py in dot(self, y, weights)
    509             result = self.clone()
    510 
--> 511         return result.mul_(y).sum()
    512 
    513     def ger(self, y):

~/.conda/envs/py38/lib/python3.8/site-packages/crypten/mpc/primitives/arithmetic.py in mul_(self, y)
    281             self.share *= y
    282             return self
--> 283         return self._arithmetic_function_(y, "mul")
    284 
    285     def div(self, y):

~/.conda/envs/py38/lib/python3.8/site-packages/crypten/mpc/primitives/arithmetic.py in _arithmetic_function_(self, y, op, *args, **kwargs)
    188 
    189     def _arithmetic_function_(self, y, op, *args, **kwargs):
--> 190         return self._arithmetic_function(y, op, inplace=True, *args, **kwargs)
    191 
    192     def _arithmetic_function(self, y, op, inplace=False, *args, **kwargs):

~/.conda/envs/py38/lib/python3.8/site-packages/crypten/mpc/primitives/arithmetic.py in _arithmetic_function(self, y, op, inplace, *args, **kwargs)
    244             else:  # scale by larger of self.encoder.scale and y.encoder.scale
    245                 if self.encoder.scale > 1 and y.encoder.scale > 1:
--> 246                     return result.div_(result.encoder.scale)
    247                 elif self.encoder.scale > 1:
    248                     result.encoder = self.encoder

~/.conda/envs/py38/lib/python3.8/site-packages/crypten/mpc/primitives/arithmetic.py in div_(self, y)
    310                 self -= wraps * 4 * (int(2 ** 62) // y)
    311             else:
--> 312                 self.share /= y
    313             return self
    314 

RuntimeError: result type Float can't be cast to the desired output type Long

[xiaolangu]

@ZJG0 Thanks! For some reason, there is no configs/default.yaml in the rebuilt version. So, I manually added this file. Anyway, it works now! Thanks for your help!

[knottb]

Closing since this seems to be resolved.