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([[8, 2, -5, 6], [8, -3, 2, 8], [0, 1, -2, -3]],
                    dtype=torch.long),
    'x': {
        'b': torch.as_tensor([[8, 1, 0], [9, 9, 7]], dtype=torch.long)
    }
})
t_shape = Size({'a': torch.Size([3, 4]), 'x': {'b': torch.Size([2, 3])}})
t_tensor = Tensor({
    'b':
    torch.as_tensor([[
        0.4011191725730896, 1.561696171760559, 0.23459962010383606,
        -0.35697418451309204
    ],
                     [
                         2.19881272315979, -0.5169054865837097,
                         -0.5217242240905762, 0.9435856938362122
                     ],
                     [
                         0.9670769572257996, -0.02175004780292511,
                         -0.9155303835868835, -1.4100911617279053
                     ]],
                    dtype=torch.float32),
    'a':
    torch.as_tensor(
        [[-1.0741915702819824, 1.4757187366485596, 0.16062210500240326],
         [-0.18332967162132263, -0.7763899564743042, 1.824703335762024]],
        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