extension-script - GitLite

KO | EN

gitlite — search

#javascript #llm #go #machine-learning #rust #deep-learning #linux #nodejs #awesome #kubernetes #vue #ios #mcp #open-source

extension-script

(★ 114)

Example repository for custom C++/CUDA operators for TorchScript

File Explorer

Download Latest Version (.zip)

- - CMakeLists.txt
  - op.cpp
- CMakeLists.txt
- main.cpp
.gitignore
Dockerfile
eager.py
example.pt
load.py
load_inline.py
README.md
script.py
setup.py
test_setup.py
trace.py

# Installation Guide

1. Get the code

git clone https://github.com/pytorch/extension-script

Downloads the entire project code from GitHub to your computer.

cd extension-script

Moves into the project folder you just downloaded.

2. Docker

Easy Recommended

Prerequisites

Git Needed to download the project code from GitHub.
Docker Desktop Needed to build and run containers. Install it and keep it running in the background.

$ docker build -t extension-script .

Builds a runnable image based on the Dockerfile.

$ docker run -v $PWD:/home -it extension-script

Runs the built image as an actual container.

✅ Run docker compose ps to check the containers are Up. If the README mentions a port, open http://localhost:PORT in your browser.

Pulled directly from this repo's README.

3. CMake

Medium

Prerequisites

Git Needed to download the project code from GitHub.
CMake The tool used to generate build configuration.
C/C++ 컴파일러 Windows needs Visual Studio (Community edition, free), macOS needs Xcode Command Line Tools, Linux needs the gcc/g++ package.

$ cmake -DCMAKE_PREFIX_PATH=/libtorch ..

Type this command into your terminal and run it.

✅ Check that an executable was created inside the build folder, then run it directly (e.g. ./build/app_name).

Pulled directly from this repo's README.

4. Python

Easy

Prerequisites

Git Needed to download the project code from GitHub.
Python 3 On Windows, be sure to check 'Add Python to PATH' during installation.

pip install .

Installs the package published on PyPI directly — no need to clone the source.

python <실행할 파일명>.py # README에서 정확한 실행 파일명을 확인하세요

Runs the Python script (or module).

✅ If it runs without errors and prints output in the terminal, it worked.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Custom TorchScript Operators Example

This repository contains examples for writing, compiling and using custom
TorchScript operators. See
[here](https://pytorch.org/tutorials/advanced/torch_script_custom_ops.html)
for the accompanying tutorial.

## Contents

There a few monuments in this repository you can visit. They are described in
context in the tutorial, which you are encouraged to read. These monuments are:

- `example_app/warp_perspective/op.cpp`: The custom operator implementation,
- `example_app/main.cpp`: An example application that loads and executes a serialized TorchScript model, which uses the custom operator, in C++,
- `script.py`: Example of using the custom operator in a scripted model,
- `trace.py`: Example of using the custom operator in a traced model,
- `eager.py`: Example of using the custom operator in vanilla eager PyTorch,
- `load.py`: Example of using `torch.utils.cpp_extension.load` to build the custom operator,
- `load.py`: Example of using `torch.utils.cpp_extension.load_inline` to build the custom operator,
- `setup.py`: Example of using setuptools to build the custom operator,
- `test_setup.py`: Example of using the custom operator built using `setup.py`.

To execute the C++ application, first run `script.py` to serialize a TorchScript
model to a file called `example.pt`, then pass that file to the
`example_app/build/example_app` binary.

## Setup

For the smoothest experience when trying out these examples, we recommend
building a docker container from this repository's `Dockerfile`. This will give
you a clean, isolated Ubuntu Linux environment in which we guarantee everything
to work perfectly. These steps should get you started:

```sh
$ git clone https://github.com/pytorch/extension-script

$ cd extension-script

$ docker build -t extension-script .

$ docker run -v $PWD:/home -it extension-script

$ root@2f00feefe46a:/home# source /activate # Activate the Conda environment

$ cd example_app && mkdir build && cd build

$ cmake -DCMAKE_PREFIX_PATH=/libtorch ..
-- The C compiler identification is GNU 5.4.0
-- The CXX compiler identification is GNU 5.4.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Looking for pthread.h
-- Looking for pthread.h - found
-- Looking for pthread_create
-- Looking for pthread_create - not found
-- Looking for pthread_create in pthreads
-- Looking for pthread_create in pthreads - not found
-- Looking for pthread_create in pthread
-- Looking for pthread_create in pthread - found
-- Found Threads: TRUE
-- Found torch: /libtorch/lib/libtorch.so
-- Configuring done
-- Generating done
-- Build files have been written to: /home/example_app/build

$ make -j
Scanning dependencies of target warp_perspective
[ 25%] Building CXX object warp_perspective/CMakeFiles/warp_perspective.dir/op.cpp.o
[ 50%] Linking CXX shared library libwarp_perspective.so
[ 50%] Built target warp_perspective
Scanning dependencies of target example_app
[ 75%] Building CXX object CMakeFiles/example_app.dir/main.cpp.o
[100%] Linking CXX executable example_app
[100%] Built target example_app
```

This will create a shared library under
`/home/example_app/build/warp_perspective/libwarp_perspective.so` containing the
custom operator defined in `example_app/warp_perspective/op.cpp`. Then, you can
run the examples, e.g.:

```sh
(base) root@2f00feefe46a:/home# python script.py
graph(%x.1 : Dynamic
      %y : Dynamic) {
  %20 : int = prim::Constant[value=1]()
  %16 : int[] = prim::Constant[value=[0, -1]]()
  %14 : int = prim::Constant[value=6]()
  %2 : int = prim::Constant[value=0]()
  %7 : int = prim::Constant[value=42]()
  %z.1 : int = prim::Constant[value=5]()
  %z.2 : int = prim::Constant[value=10]()
  %13 : int = prim::Constant[value=3]()
  %4 : Dynamic = aten::select(%x.1, %2, %2)
  %6 : Dynamic = aten::select(%4, %2, %2)
  %8 : Dynamic = aten::eq(%6, %7)
  %9 : bool = prim::TensorToBool(%8)
  %z : int = prim::If(%9)
    block0() {
      -> (%z.1)
    }
    block1() {
      -> (%z.2)
    }
  %17 : Dynamic = aten::eye(%13, %14, %2, %16)
  %x : Dynamic = my_ops::warp_perspective(%x.1, %17)
  %19 : Dynamic = aten::matmul(%x, %y)
  %21 : Dynamic = aten::add(%19, %z, %20)
  return (%21);
}

tensor([[11.6196, 12.0056, 11.6122, 12.9298,  7.0649],
        [ 8.5063,  9.0621,  9.9925,  6.3741,  8.9668],
        [12.5898,  6.5872,  8.1511, 10.0806, 11.9829],
        [ 4.9142, 11.6614, 15.7161, 17.0538, 11.7243],
        [10.0000, 10.0000, 10.0000, 10.0000, 10.0000],
        [10.0000, 10.0000, 10.0000, 10.0000, 10.0000],
        [10.0000, 10.0000, 10.0000, 10.0000, 10.0000],
        [10.0000, 10.0000, 10.0000, 10.0000, 10.0000]])
```