# torch_link

**Repository Path**: lite_xay/torch_link

## Basic Information

- **Project Name**: torch_link
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2024-03-20
- **Last Updated**: 2024-03-26

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Torch Link

A component to link torch and mindspore_lite. User can utilize it to convert torch.Tensor to mindspore_lite.Tensor or mindspore_lite.Tensor to torch.Tensor without copying of data.

> The owner of data is src Tensor, dst Tensor only has the right to use the data.

## How to compile

### Compilation dependencies of Cpp:
- [GCC](https://gcc.gnu.org/releases.html) >= 7.3.0
- CMake](https://cmake.org/download/) >= 3.18.3

### Compilation dependencies of Python:
- [Python](https://www.python.org/) >= 3.7.0
- [torch](https://pytorch.org/) >= 2.1.0
- [mindspore-lite](https://www.mindspore.cn/lite/docs/zh-CN/r2.2/use/downloads.html) >= 2.2

### Instructions for parameters of `build.sh`

| Parameter | Parameter Description                                        |
| --------- | ------------------------------------------------------------ |
| -h        | Print usage.                                                 |
| -v        | Control to show the details of compilation.                  |
| -d        | Open Debug mode when compiling. Default is Release mode.     |
| -i        | Control if auto-install after compilation. Default only compilation. |
| -j        | Set the number of threads used to compile. Default is 4      |

### Module build compilation options

| Option                | Parameter Description                               | Necessary |
| --------------------- | --------------------------------------------------- | --------- |
| MSLITE_INSTALLED_PATH | Env of the installation path for mindspore_lite     | True      |
| TORCH_INSTALLED_PATH  | Env of the installation path for torch              | True      |
| ENABLE_GITEE_MIRROR   | Control the download source of dependency libraries | False     |

### Compilation

1. set the necessary

```bash
export MSLITE_INSTALL_PATH=xxx
export TORCH_INSTALL_PATH=xxx
export ENABLE_GITEE_MIRROR=on
```

2. run `build.sh`
    ```bash
    bash build.sh -i
    ```
## Usage

### Description of API

- Convert torch.Tensor to mindspore_lite.Tensor
  ```python
  def torch_tensor_2_mslite_tensor(tensor: torch.Tensor) -> mslite.Tensor
  ```

- Convert mindspore_lite.Tensor to torch.Tensor
  ```python
  def mslite_tensor_2_torch_tensor(tensor: torch.Tensor) -> mslite.Tensor
  ```

### Example

- Convert torch.Tensor to mindspore_lite.Tensor
  ```python
  import torch
  import torch_link
  import numpy as np

  np_data = np.random.uniform(size=[2, 3]).astype(np.float32)
  torch_tensor = torch.from_numpy(np_data)
  mslite_tensor = torch_link.torch_tensor_2_mslite_tensor(torch_tensor)
  ```

- Convert mindspore_lite.Tensor to torch.Tensor
  ```python
  import mindspore_lite as mslite
  import torch_link
  import numpy as np

  np_data = np.random.uniform(size=[2, 3]).astype(np.float32)
  mslite_tensor = mslite.Tensor(np_data)
  torch_tensor = torch_link.mslite_tensor_2_torch_tensor(mslite_tensor)
  ```