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([[1, -2, -3], [9, 5, 0]], dtype=torch.long) }, 'a': torch.as_tensor([[2, 5, 3, 1], [-4, -3, 9, -5], [7, 6, 4, -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([[ -0.6771437525749207, 0.6759194135665894, 0.055163294076919556, -0.3675537109375 ], [ 1.3639285564422607, -1.1286232471466064, 1.8531461954116821, 0.35475650429725647 ], [ -0.5053892135620117, 1.1922693252563477, 0.5875821113586426, 1.1067378520965576 ]], dtype=torch.float32), 'a': torch.as_tensor( [[-1.4010660648345947, -0.28158828616142273, 0.545931875705719], [0.1472681611776352, -0.5047093629837036, 0.8301571607589722]], 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