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([[6, -5, 0, 7], [-1, 4, 9, 9], [7, -3, 0, -5]], dtype=torch.long), 'x': { 'b': torch.as_tensor([[1, 1, 3], [1, 8, 0]], dtype=torch.long) } }) t_shape = Size({'a': torch.Size([3, 4]), 'x': {'b': torch.Size([2, 3])}}) t_tensor = Tensor({ 'a': torch.as_tensor( [[-0.7283960580825806, -1.1181594133377075, -0.39858317375183105], [-0.44914159178733826, -0.6912498474121094, 1.0196959972381592]], dtype=torch.float32), 'b': torch.as_tensor([[ 0.7337965965270996, -0.33806368708610535, -1.2548998594284058, 0.12338656932115555 ], [ 1.2900900840759277, -0.22137492895126343, -2.83416748046875, 0.9930022358894348 ], [ -1.5423628091812134, 0.1453082412481308, 0.52750563621521, -0.2628335654735565 ]], 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