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([[2, 6, -2], [-1, 1, 0]], dtype=torch.long) }, 'a': torch.as_tensor([[8, -5, -5, -1], [6, 3, 4, 4], [-2, 9, -3, 6]], dtype=torch.long) }) t_shape = Size({'x': {'b': torch.Size([2, 3])}, 'a': torch.Size([3, 4])}) t_tensor = Tensor({ 'a': torch.as_tensor( [[0.25546881556510925, 1.5697441101074219, 0.6671493649482727], [0.29678308963775635, 0.5242627859115601, 0.8216266632080078]], dtype=torch.float32), 'b': torch.as_tensor([[ 1.484199047088623, 0.22663825750350952, 1.7510405778884888, -0.40903058648109436 ], [ -0.6381224393844604, 1.3796913623809814, 0.6873614192008972, 0.07700836658477783 ], [ -0.02083461917936802, -0.9499525427818298, 0.4672611653804779, 0.04375791922211647 ]], 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