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([[-4, 1, -1], [5, -1, 1]], dtype=torch.long) }, 'a': torch.as_tensor([[4, 7, 6, -3], [1, 9, 8, -4], [1, 4, -1, -5]], dtype=torch.long) }) t_shape = Size({'x': {'b': torch.Size([2, 3])}, 'a': torch.Size([3, 4])}) t_tensor = Tensor({ 'b': torch.as_tensor([[ 0.46654418110847473, -0.2631285786628723, 0.2108316570520401, 1.036888599395752 ], [ 1.0321507453918457, -1.731726884841919, -0.6699100732803345, -0.7738327383995056 ], [ -1.658908724784851, 1.4187813997268677, 2.4288148880004883, -0.4278255105018616 ]], dtype=torch.float32), 'a': torch.as_tensor( [[-0.4673236906528473, -0.9881297945976257, 0.42380189895629883], [2.3155558109283447, 0.24167941510677338, 0.6736022233963013]], 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