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, 2, -5, 6], [8, -3, 2, 8], [0, 1, -2, -3]], dtype=torch.long), 'x': { 'b': torch.as_tensor([[8, 1, 0], [9, 9, 7]], dtype=torch.long) } }) t_shape = Size({'a': torch.Size([3, 4]), 'x': {'b': torch.Size([2, 3])}}) t_tensor = Tensor({ 'b': torch.as_tensor([[ 0.4011191725730896, 1.561696171760559, 0.23459962010383606, -0.35697418451309204 ], [ 2.19881272315979, -0.5169054865837097, -0.5217242240905762, 0.9435856938362122 ], [ 0.9670769572257996, -0.02175004780292511, -0.9155303835868835, -1.4100911617279053 ]], dtype=torch.float32), 'a': torch.as_tensor( [[-1.0741915702819824, 1.4757187366485596, 0.16062210500240326], [-0.18332967162132263, -0.7763899564743042, 1.824703335762024]], 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