Plugins

Potc support

Potc is a package that can convert any object into executable source code. For DI-treetensor, potc can support the source code transformation of treevalue objects through the installation of additional plugins. So we can execute the following installation command

pip install DI-treetensor[potc]

After this installation, you will be able to directly convert tree-nested tensors to objects without any additional operations. Such as

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
from potc import transvars

import treetensor.torch as ttorch

t_tensor = ttorch.randn({'a': (2, 3), 'b': (3, 4)})
t_i_tensor = ttorch.randint(-5, 10, {'a': (3, 4), 'x': {'b': (2, 3)}})
t_shape = t_i_tensor.shape

if __name__ == '__main__':
    _code = transvars(
        {
            't_tensor': t_tensor,
            't_i_tensor': t_i_tensor,
            't_shape': t_shape,
        },
        reformat='pep8'
    )
    print(_code)

The output should be

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
import torch
from treetensor import Tensor
from treetensor.torch import Size

__all__ = ['t_i_tensor', 't_shape', 't_tensor']
t_i_tensor = Tensor({
    'a':
    torch.as_tensor([[-2, 0, 4, 2], [-1, 2, 6, 6], [-5, -1, 6, 5]],
                    dtype=torch.long),
    'x': {
        'b': torch.as_tensor([[-5, -1, -4], [5, -2, -5]], dtype=torch.long)
    }
})
t_shape = Size({'a': torch.Size([3, 4]), 'x': {'b': torch.Size([2, 3])}})
t_tensor = Tensor({
    'b':
    torch.as_tensor([[
        -0.15807105600833893, 0.9869658946990967, 1.396366000175476,
        0.36078885197639465
    ],
                     [
                         0.4816668629646301, 1.1220568418502808,
                         -0.7234281301498413, -0.2238752543926239
                     ],
                     [
                         -0.21661117672920227, -0.6312008500099182,
                         -1.4726037979125977, -1.2045934200286865
                     ]],
                    dtype=torch.float32),
    'a':
    torch.as_tensor(
        [[2.203050136566162, 0.3990941047668457, -1.1317102909088135],
         [0.34639817476272583, -0.13156738877296448, 0.014780695550143719]],
        dtype=torch.float32)
})

Also, you can use the following CLI command to get the same output results as above.

potc export -v 'test_simple.t_*'

For further information, you can refer to