reproduce.py

import torch

self_tensors = [torch.rand(2, 3, requires_grad=True),
                torch.rand(4, 5, requires_grad=True)]

self_tensors = [torch.rand(2, 3),
                torch.rand(4, 5)]

flat_tensors = []
for tensor in self_tensors:
    flat_tensors.append(tensor.view(-1))
buffer_ = torch.cat(flat_tensors)
current_offset = 0
for i in range(len(self_tensors)):
    self_tensors[i].set_(buffer_.storage(),
                    storage_offset=current_offset,
                    size=self_tensors[i].size(),
                    stride=self_tensors[i].stride())
    current_offset += self_tensors[i].numel()

print('self_tensors 1')
print(self_tensors)
buffer_.add_(2)
print('self_tensors 2')
print(self_tensors)