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([[5, 0, 6, 7], [3, 5, -5, -5], [1, 2, 9, 4]], dtype=torch.long), 'x': { 'b': torch.as_tensor([[-4, 8, 8], [-4, 9, -5]], dtype=torch.long) } }) t_shape = Size({'a': torch.Size([3, 4]), 'x': {'b': torch.Size([2, 3])}}) t_tensor = Tensor({ 'a': torch.as_tensor( [[1.0884815454483032, 1.1385016441345215, 0.29884859919548035], [-0.20478612184524536, -1.4196021556854248, -0.04051636531949043]], dtype=torch.float32), 'b': torch.as_tensor([[ 2.595019578933716, 0.6456472277641296, 1.2577778100967407, 2.0272738933563232 ], [ 1.150585412979126, 1.185786485671997, -1.8594878911972046, 0.02676238678395748 ], [ 2.1013214588165283, -1.028666377067566, 0.758303701877594, -0.3191949725151062 ]], 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