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({
    'x': {
        'b': torch.as_tensor([[-4, 0, -5], [5, 4, 5]], dtype=torch.long)
    },
    'a':
    torch.as_tensor([[8, -2, -3, -1], [9, -1, 2, 0], [-4, 0, -1, 6]],
                    dtype=torch.long)
})
t_shape = Size({'x': {'b': torch.Size([2, 3])}, 'a': torch.Size([3, 4])})
t_tensor = Tensor({
    'b':
    torch.as_tensor([[
        1.136405110359192, -0.9269028306007385, 1.177195429801941,
        0.6669424176216125
    ],
                     [
                         0.5716557502746582, -0.5980654954910278,
                         -1.3264665603637695, -0.35822397470474243
                     ],
                     [
                         -0.9364849925041199, -0.13343240320682526,
                         1.6067432165145874, 1.1122426986694336
                     ]],
                    dtype=torch.float32),
    'a':
    torch.as_tensor(
        [[-0.5386978983879089, -0.5760756731033325, -0.7756832838058472],
         [0.6613494753837585, -1.534128189086914, 1.44227135181427]],
        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