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, -3, 5, -5], [1, 5, 6, 5], [0, 8, 1, -4]],
                    dtype=torch.long),
    'x': {
        'b': torch.as_tensor([[0, 3, 8], [-3, -3, -3]], dtype=torch.long)
    }
})
t_shape = Size({'a': torch.Size([3, 4]), 'x': {'b': torch.Size([2, 3])}})
t_tensor = Tensor({
    'a':
    torch.as_tensor(
        [[-2.4147095680236816, -1.3996179103851318, 0.3234967887401581],
         [1.2817164659500122, -1.338171124458313, -0.7192679643630981]],
        dtype=torch.float32),
    'b':
    torch.as_tensor([[
        0.839584231376648, -2.199462413787842, -0.04107382893562317,
        0.0770263671875
    ],
                     [
                         -1.587693214416504, -0.8288155794143677,
                         -0.6327997446060181, -0.07201468199491501
                     ],
                     [
                         1.6985743045806885, -1.0063565969467163,
                         0.5097642540931702, -2.029224157333374
                     ]],
                    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