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([[-2, -3, 5, -5], [1, 5, 6, 5], [0, 8, 1, -4]], dtype=torch.long), 'x': { 'b': torch.as_tensor([[0, 3, 8], [-3, -3, -3]], dtype=torch.long) } }) t_shape = Size({'a': torch.Size([3, 4]), 'x': {'b': torch.Size([2, 3])}}) t_tensor = Tensor({ 'a': torch.as_tensor( [[-2.4147095680236816, -1.3996179103851318, 0.3234967887401581], [1.2817164659500122, -1.338171124458313, -0.7192679643630981]], dtype=torch.float32), 'b': torch.as_tensor([[ 0.839584231376648, -2.199462413787842, -0.04107382893562317, 0.0770263671875 ], [ -1.587693214416504, -0.8288155794143677, -0.6327997446060181, -0.07201468199491501 ], [ 1.6985743045806885, -1.0063565969467163, 0.5097642540931702, -2.029224157333374 ]], 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