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, 6, 4, 4], [4, 5, 9, -1], [4, -5, 6, -1]], dtype=torch.long), 'x': { 'b': torch.as_tensor([[-2, -1, 7], [5, 6, 4]], dtype=torch.long) } }) t_shape = Size({'a': torch.Size([3, 4]), 'x': {'b': torch.Size([2, 3])}}) t_tensor = Tensor({ 'a': torch.as_tensor( [[0.38337624073028564, 1.7304457426071167, 0.08502542972564697], [-0.4015350341796875, 0.6952739953994751, -0.827825129032135]], dtype=torch.float32), 'b': torch.as_tensor([[ 0.6358466148376465, -1.2224129438400269, 0.8938352465629578, -1.4846575260162354 ], [ -0.007553907111287117, -1.3188650608062744, -1.1571229696273804, -1.2482050657272339 ], [ 1.3722939491271973, -0.8350231051445007, 0.6051339507102966, 0.8400365114212036 ]], 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