potc-torch

Potc support for torch module.

Installation

You can simply install it with pip command line from the official PyPI site.

pip install potc-torch

Quick Start

After potc-torch is installed, you can convert the torch objects into executable source code without any additional operations.

We can create a python script which is named test_simple.py

import torch
from potc import transvars

a = torch.randn(2, 3)
b = torch.randint(-5, 10, (3, 4))
bs = b.size()
bd = b.device
if __name__ == '__main__':
    _code = transvars(
        {'a': a, 'b': b, 'bs': bs, 'bd': bd},
        reformat='pep8',
    )
    print(_code)

The output result should be like this (may be slightly different because of the usage of torch.randn and torch.randint)

import torch

__all__ = ['a', 'b', 'bd', 'bs']
a = torch.as_tensor(
    [[0.6224261522293091, 0.4725508689880371, 0.45328783988952637],
     [-0.5855962634086609, 0.4898407459259033, 0.4769541621208191]],
    dtype=torch.float32)
b = torch.as_tensor([[2, 7, 4, -3], [-2, 4, 8, 1], [7, -5, 3, 6]],
                    dtype=torch.long)
bd = torch.device('cpu')
bs = torch.Size([3, 4])

And, you can use the following CLI command to get the same output results as above.

potc export -v 'test_simple.a' -v 'test_simple.b*'

Contributing

We appreciate all contributions to improve potc and potc-torch, both logic and system designs. Please refer to CONTRIBUTING.md for more guides.

License

potc-torch released under the Apache 2.0 license.