pytorch_dart #

A dart wrapper for Libtorch,provides some fundametal bindings.

Getting Started #

1. Add pytorch_dart to your pubspec.yaml

2.  dart run pytorch_dart:setup
   

3.  import 'package:pytorch_dart/pytorch_dart.dart' as torch;
   
   

Usage #

1. It only include some basic functions about torch.Tensor now
2. Dart code：

   torch.Tensor tensor = torch.createTensorFromList(
     [[ 1.0028, -0.9893,  0.5809],
       [-0.1669,  0.7299,  0.4942]]
   );
   List<dynamic> listOrig = torch.createListFromTensor(tensor);
   torch.Tensor t = tensor.transpose( 0, 1);
   List<dynamic> listTranspose = torch.createListFromTensor(t);
   
   torch.Tensor t1 = torch.createTensorFromList([
     [1.1, 2.2],
     [3.3, 4.4]
   ]);
   torch.Tensor t2 = torch.createTensorFromList([
     [0.1, 0.2],
     [0.3, 0.4]
   ]);
   torch.Tensor t3 = torch.sub(t1, t2);
   torch.Tensor t4 = torch.add(t1, t2);
   List<dynamic> list1 = torch.createListFromTensor(t1);
   List<dynamic> list2 = torch.createListFromTensor(t2);
   List<dynamic> list3 = torch.createListFromTensor(t3);
   List<dynamic> list4 = torch.createListFromTensor(t4);
   print("original Tensor=$listOrig");
   print("transposed Tensor=$listTranspose");
   print("Tensor t1=$list1");
   print("Tensor t2=$list2");
   print("Tensor t3=t1-t2=$list3");
   print("Tensor t4=t1+t2=$list4");
   
   

3. Corresponding python code:

import torch;
tensor=torch.Tensor([[ 1.0028, -0.9893,  0.5809],
        [-0.1669,  0.7299,  0.4942]])
t=tensor.transpose(0,1)
t1=torch.Tensor([
      [1.1, 2.2],
      [3.3, 4.4]
    ])
t2=torch.Tensor([
      [0.1, 0.2],
      [0.3, 0.4]
    ])
t3=torch.sub(t1,t2)
t4=torch.add(t1,t2)
print(t1)
print(t2)
print(t3)
print(t4)

Roadmap #

1. Add support for other functions such as torch.permute()
2. Add support for Windows
3. Add support for other functions,such as torch.nn

Acknowledgement #

This project uses pytorch-flutter-FFI-example