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, 1, -1], [5, -1, 1]], dtype=torch.long)
    },
    'a':
    torch.as_tensor([[4, 7, 6, -3], [1, 9, 8, -4], [1, 4, -1, -5]],
                    dtype=torch.long)
})
t_shape = Size({'x': {'b': torch.Size([2, 3])}, 'a': torch.Size([3, 4])})
t_tensor = Tensor({
    'b':
    torch.as_tensor([[
        0.46654418110847473, -0.2631285786628723, 0.2108316570520401,
        1.036888599395752
    ],
                     [
                         1.0321507453918457, -1.731726884841919,
                         -0.6699100732803345, -0.7738327383995056
                     ],
                     [
                         -1.658908724784851, 1.4187813997268677,
                         2.4288148880004883, -0.4278255105018616
                     ]],
                    dtype=torch.float32),
    'a':
    torch.as_tensor(
        [[-0.4673236906528473, -0.9881297945976257, 0.42380189895629883],
         [2.3155558109283447, 0.24167941510677338, 0.6736022233963013]],
        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