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, 6, -2], [5, 2, 7]], dtype=torch.long) }, 'a': torch.as_tensor([[4, 1, -1, 0], [0, 2, 3, 8], [9, 5, 0, 4]], dtype=torch.long) }) t_shape = Size({'x': {'b': torch.Size([2, 3])}, 'a': torch.Size([3, 4])}) t_tensor = Tensor({ 'a': torch.as_tensor( [[1.6284724473953247, 0.3768380582332611, 0.6547043919563293], [-0.8493189215660095, 0.24233856797218323, -0.3979470729827881]], dtype=torch.float32), 'b': torch.as_tensor([[ 0.1390693634748459, 2.363795042037964, -1.0959696769714355, -1.8152844905853271 ], [ -0.464792400598526, -0.10408748686313629, 0.3499223291873932, -0.19817695021629333 ], [ 3.5152814388275146, -2.1466362476348877, -0.11760104447603226, -0.3095119893550873 ]], 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