扩展 TorchScript 以使用自定义 C++类
创建于:2025 年 4 月 1 日 | 最后更新:2025 年 4 月 1 日 | 最后验证:2024 年 11 月 5 日
警告
TorchScript 不再处于活跃开发状态。
本教程是自定义操作符教程的后续教程,介绍了我们构建的用于将 C++类绑定到 TorchScript 和 Python 的 API。该 API 与 pybind11 非常相似,如果您熟悉该系统,大多数概念都将得以迁移。
在 C++中实现和绑定类
在本教程中,我们将定义一个简单的 C++类,该类在成员变量中维护持久状态。
// This header is all you need to do the C++ portions of this
// tutorial
#include <torch/script.h>
// This header is what defines the custom class registration
// behavior specifically. script.h already includes this, but
// we include it here so you know it exists in case you want
// to look at the API or implementation.
#include <torch/custom_class.h>
#include <string>
#include <vector>
template <class T>
struct MyStackClass : torch::CustomClassHolder {
std::vector<T> stack_;
MyStackClass(std::vector<T> init) : stack_(init.begin(), init.end()) {}
void push(T x) {
stack_.push_back(x);
}
T pop() {
auto val = stack_.back();
stack_.pop_back();
return val;
}
c10::intrusive_ptr<MyStackClass> clone() const {
return c10::make_intrusive<MyStackClass>(stack_);
}
void merge(const c10::intrusive_ptr<MyStackClass>& c) {
for (auto& elem : c->stack_) {
push(elem);
}
}
};
有几点需要注意:
torch/custom_class.h
是您需要包含以扩展 TorchScript 的自定义类的头文件。注意,每当我们在处理自定义类的实例时,我们都是通过实例化的
c10::intrusive_ptr<>
来进行的。将intrusive_ptr
视为一个类似于std::shared_ptr
的智能指针,但引用计数是直接存储在对象中的,而不是像std::shared_ptr
那样存储在单独的元数据块中。torch::Tensor
内部使用相同的指针类型;并且自定义类也必须使用这种指针类型,这样我们才能一致地管理不同的对象类型。第二点要注意的是,用户定义的类必须继承自
torch::CustomClassHolder
。这确保了自定义类有空间来存储引用计数。
现在我们来看看如何使这个类对 TorchScript 可见,这个过程称为绑定类:
// Notice a few things:
// - We pass the class to be registered as a template parameter to
// `torch::class_`. In this instance, we've passed the
// specialization of the MyStackClass class ``MyStackClass<std::string>``.
// In general, you cannot register a non-specialized template
// class. For non-templated classes, you can just pass the
// class name directly as the template parameter.
// - The arguments passed to the constructor make up the "qualified name"
// of the class. In this case, the registered class will appear in
// Python and C++ as `torch.classes.my_classes.MyStackClass`. We call
// the first argument the "namespace" and the second argument the
// actual class name.
TORCH_LIBRARY(my_classes, m) {
m.class_<MyStackClass<std::string>>("MyStackClass")
// The following line registers the contructor of our MyStackClass
// class that takes a single `std::vector<std::string>` argument,
// i.e. it exposes the C++ method `MyStackClass(std::vector<T> init)`.
// Currently, we do not support registering overloaded
// constructors, so for now you can only `def()` one instance of
// `torch::init`.
.def(torch::init<std::vector<std::string>>())
// The next line registers a stateless (i.e. no captures) C++ lambda
// function as a method. Note that a lambda function must take a
// `c10::intrusive_ptr<YourClass>` (or some const/ref version of that)
// as the first argument. Other arguments can be whatever you want.
.def("top", [](const c10::intrusive_ptr<MyStackClass<std::string>>& self) {
return self->stack_.back();
})
// The following four lines expose methods of the MyStackClass<std::string>
// class as-is. `torch::class_` will automatically examine the
// argument and return types of the passed-in method pointers and
// expose these to Python and TorchScript accordingly. Finally, notice
// that we must take the *address* of the fully-qualified method name,
// i.e. use the unary `&` operator, due to C++ typing rules.
.def("push", &MyStackClass<std::string>::push)
.def("pop", &MyStackClass<std::string>::pop)
.def("clone", &MyStackClass<std::string>::clone)
.def("merge", &MyStackClass<std::string>::merge)
;
}
使用 CMake 构建 C++ 项目示例 ¶
现在,我们将使用 CMake 构建系统构建上述 C++代码。首先,将到目前为止我们所涵盖的所有 C++代码放入一个名为 class.cpp
的文件中。然后,编写一个简单的 CMakeLists.txt
文件并将其放入同一目录下。 CMakeLists.txt
应该看起来像这样:
cmake_minimum_required(VERSION 3.1 FATAL_ERROR)
project(custom_class)
find_package(Torch REQUIRED)
# Define our library target
add_library(custom_class SHARED class.cpp)
set(CMAKE_CXX_STANDARD 14)
# Link against LibTorch
target_link_libraries(custom_class "${TORCH_LIBRARIES}")
此外,创建一个 build
目录。你的文件结构应该如下所示:
custom_class_project/
class.cpp
CMakeLists.txt
build/
我们假设你已经按照前面教程中的描述设置了你的环境。接下来,运行 cmake 然后 make 来构建项目:
$ cd build
$ cmake -DCMAKE_PREFIX_PATH="$(python -c 'import torch.utils; print(torch.utils.cmake_prefix_path)')" ..
-- The C compiler identification is GNU 7.3.1
-- The CXX compiler identification is GNU 7.3.1
-- Check for working C compiler: /opt/rh/devtoolset-7/root/usr/bin/cc
-- Check for working C compiler: /opt/rh/devtoolset-7/root/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: /opt/rh/devtoolset-7/root/usr/bin/c++
-- Check for working CXX compiler: /opt/rh/devtoolset-7/root/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: /torchbind_tutorial/libtorch/lib/libtorch.so
-- Configuring done
-- Generating done
-- Build files have been written to: /torchbind_tutorial/build
$ make -j
Scanning dependencies of target custom_class
[ 50%] Building CXX object CMakeFiles/custom_class.dir/class.cpp.o
[100%] Linking CXX shared library libcustom_class.so
[100%] Built target custom_class
你会发现构建目录中现在(以及其他一些东西)有一个动态库文件。在 Linux 上,这个文件可能命名为 libcustom_class.so
。所以文件结构应该如下所示:
custom_class_project/
class.cpp
CMakeLists.txt
build/
libcustom_class.so
在 Python 中使用 C++类和 TorchScript ¶
现在我们已经将我们的类及其注册编译成了 .so
文件,我们可以将那个.so 文件加载到 Python 中并尝试运行。以下是一个演示该过程的脚本:
import torch
# `torch.classes.load_library()` allows you to pass the path to your .so file
# to load it in and make the custom C++ classes available to both Python and
# TorchScript
torch.classes.load_library("build/libcustom_class.so")
# You can query the loaded libraries like this:
print(torch.classes.loaded_libraries)
# prints {'/custom_class_project/build/libcustom_class.so'}
# We can find and instantiate our custom C++ class in python by using the
# `torch.classes` namespace:
#
# This instantiation will invoke the MyStackClass(std::vector<T> init)
# constructor we registered earlier
s = torch.classes.my_classes.MyStackClass(["foo", "bar"])
# We can call methods in Python
s.push("pushed")
assert s.pop() == "pushed"
# Test custom operator
s.push("pushed")
torch.ops.my_classes.manipulate_instance(s) # acting as s.pop()
assert s.top() == "bar"
# Returning and passing instances of custom classes works as you'd expect
s2 = s.clone()
s.merge(s2)
for expected in ["bar", "foo", "bar", "foo"]:
assert s.pop() == expected
# We can also use the class in TorchScript
# For now, we need to assign the class's type to a local in order to
# annotate the type on the TorchScript function. This may change
# in the future.
MyStackClass = torch.classes.my_classes.MyStackClass
@torch.jit.script
def do_stacks(s: MyStackClass): # We can pass a custom class instance
# We can instantiate the class
s2 = torch.classes.my_classes.MyStackClass(["hi", "mom"])
s2.merge(s) # We can call a method on the class
# We can also return instances of the class
# from TorchScript function/methods
return s2.clone(), s2.top()
stack, top = do_stacks(torch.classes.my_classes.MyStackClass(["wow"]))
assert top == "wow"
for expected in ["wow", "mom", "hi"]:
assert stack.pop() == expected
使用自定义类保存、加载和运行 TorchScript 代码 ¶
我们还可以在 libtorch 的帮助下,在 C++进程中使用自定义注册的 C++类。作为一个例子,让我们定义一个简单的 nn.Module
,它实例化并调用我们的 MyStackClass 类的方法:
import torch
torch.classes.load_library('build/libcustom_class.so')
class Foo(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, s: str) -> str:
stack = torch.classes.my_classes.MyStackClass(["hi", "mom"])
return stack.pop() + s
scripted_foo = torch.jit.script(Foo())
print(scripted_foo.graph)
scripted_foo.save('foo.pt')
我们文件系统中现在包含了我们刚刚定义的序列化 TorchScript 程序。
现在,我们将定义一个新的 CMake 项目,以展示如何加载此模型及其所需的.so 文件。有关如何操作的完整说明,请参阅 C++中加载 TorchScript 模型教程。
与之前类似,让我们创建一个包含以下内容的文件结构:
cpp_inference_example/
infer.cpp
CMakeLists.txt
foo.pt
build/
custom_class_project/
class.cpp
CMakeLists.txt
build/
注意我们已经复制了序列化的 foo.pt
文件,以及来自 custom_class_project
的源树。我们将把 custom_class_project
作为依赖项添加到这个 C++项目中,以便将自定义类构建到二进制文件中。
让我们将以下内容填充到 infer.cpp
中:
#include <torch/script.h>
#include <iostream>
#include <memory>
int main(int argc, const char* argv[]) {
torch::jit::Module module;
try {
// Deserialize the ScriptModule from a file using torch::jit::load().
module = torch::jit::load("foo.pt");
}
catch (const c10::Error& e) {
std::cerr << "error loading the model\n";
return -1;
}
std::vector<c10::IValue> inputs = {"foobarbaz"};
auto output = module.forward(inputs).toString();
std::cout << output->string() << std::endl;
}
类似地,让我们定义我们的 CMakeLists.txt 文件:
cmake_minimum_required(VERSION 3.1 FATAL_ERROR)
project(infer)
find_package(Torch REQUIRED)
add_subdirectory(custom_class_project)
# Define our library target
add_executable(infer infer.cpp)
set(CMAKE_CXX_STANDARD 14)
# Link against LibTorch
target_link_libraries(infer "${TORCH_LIBRARIES}")
# This is where we link in our libcustom_class code, making our
# custom class available in our binary.
target_link_libraries(infer -Wl,--no-as-needed custom_class)
你知道该怎么做: cd build
, cmake
,和 make
:
$ cd build
$ cmake -DCMAKE_PREFIX_PATH="$(python -c 'import torch.utils; print(torch.utils.cmake_prefix_path)')" ..
-- The C compiler identification is GNU 7.3.1
-- The CXX compiler identification is GNU 7.3.1
-- Check for working C compiler: /opt/rh/devtoolset-7/root/usr/bin/cc
-- Check for working C compiler: /opt/rh/devtoolset-7/root/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: /opt/rh/devtoolset-7/root/usr/bin/c++
-- Check for working CXX compiler: /opt/rh/devtoolset-7/root/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: /local/miniconda3/lib/python3.7/site-packages/torch/lib/libtorch.so
-- Configuring done
-- Generating done
-- Build files have been written to: /cpp_inference_example/build
$ make -j
Scanning dependencies of target custom_class
[ 25%] Building CXX object custom_class_project/CMakeFiles/custom_class.dir/class.cpp.o
[ 50%] Linking CXX shared library libcustom_class.so
[ 50%] Built target custom_class
Scanning dependencies of target infer
[ 75%] Building CXX object CMakeFiles/infer.dir/infer.cpp.o
[100%] Linking CXX executable infer
[100%] Built target infer
现在我们可以运行我们的激动人心的 C++可执行文件了:
$ ./infer
momfoobarbaz
太棒了!
将自定义类移动到/从 IValues
也可能需要将自定义类从自定义 C++ 类实例移动到或移出 IValue``s, such as when you take or return ``IValue``s from TorchScript methods
or you want to instantiate a custom class attribute in C++. For creating an
``IValue
:
torch::make_custom_class<T>()
提供了一个类似于 c10::intrusive_ptr 的 API,它会接受你提供的任何参数集,调用匹配该参数集的 T 的构造函数,并将该实例包装起来并返回。然而,它返回的不是指向自定义类对象的指针,而是返回一个包含对象的IValue
。然后你可以直接将这个IValue
传递给 TorchScript。如果您已经有一个指向您的类的
intrusive_ptr
,您可以直接使用构造函数IValue(intrusive_ptr<T>)
从它构造一个 IValue。
对于将 IValue
转换回自定义类:
IValue::toCustomClass<T>()
将返回一个指向IValue
包含的自定义类的intrusive_ptr<T>
。内部,此函数会检查T
是否已注册为自定义类,并且IValue
确实包含一个自定义类。您可以通过调用isCustomClass()
手动检查IValue
是否包含自定义类。
定义自定义 C++类的序列化/反序列化方法
如果你尝试将自定义绑定的 C++类作为属性保存 ScriptModule
,你会得到以下错误:
# export_attr.py
import torch
torch.classes.load_library('build/libcustom_class.so')
class Foo(torch.nn.Module):
def __init__(self):
super().__init__()
self.stack = torch.classes.my_classes.MyStackClass(["just", "testing"])
def forward(self, s: str) -> str:
return self.stack.pop() + s
scripted_foo = torch.jit.script(Foo())
scripted_foo.save('foo.pt')
loaded = torch.jit.load('foo.pt')
print(loaded.stack.pop())
$ python export_attr.py
RuntimeError: Cannot serialize custom bound C++ class __torch__.torch.classes.my_classes.MyStackClass. Please define serialization methods via def_pickle for this class. (pushIValueImpl at ../torch/csrc/jit/pickler.cpp:128)
这是因为 TorchScript 无法自动确定从您的 C++类中保存的信息。您必须手动指定。这样做的方法是在类上使用 class_
的特殊 def_pickle
方法定义 __getstate__
和 __setstate__
方法。
备注
在 TorchScript 中, __getstate__
和 __setstate__
的语义与 Python pickle 模块的语义相同。你可以了解更多关于我们如何使用这些方法的信息。
下面是一个示例,我们可以添加到 MyStackClass
的注册中,以包括序列化方法:
// class_<>::def_pickle allows you to define the serialization
// and deserialization methods for your C++ class.
// Currently, we only support passing stateless lambda functions
// as arguments to def_pickle
.def_pickle(
// __getstate__
// This function defines what data structure should be produced
// when we serialize an instance of this class. The function
// must take a single `self` argument, which is an intrusive_ptr
// to the instance of the object. The function can return
// any type that is supported as a return value of the TorchScript
// custom operator API. In this instance, we've chosen to return
// a std::vector<std::string> as the salient data to preserve
// from the class.
[](const c10::intrusive_ptr<MyStackClass<std::string>>& self)
-> std::vector<std::string> {
return self->stack_;
},
// __setstate__
// This function defines how to create a new instance of the C++
// class when we are deserializing. The function must take a
// single argument of the same type as the return value of
// `__getstate__`. The function must return an intrusive_ptr
// to a new instance of the C++ class, initialized however
// you would like given the serialized state.
[](std::vector<std::string> state)
-> c10::intrusive_ptr<MyStackClass<std::string>> {
// A convenient way to instantiate an object and get an
// intrusive_ptr to it is via `make_intrusive`. We use
// that here to allocate an instance of MyStackClass<std::string>
// and call the single-argument std::vector<std::string>
// constructor with the serialized state.
return c10::make_intrusive<MyStackClass<std::string>>(std::move(state));
});
备注
我们在 pickle API 中采取了与 pybind11 不同的方法。pybind11 作为一个特殊函数 pybind11::pickle()
,您将其传递给 class_::def()
,而我们为此目的有一个单独的方法 def_pickle
。这是因为名称 torch::jit::pickle
已经被占用,我们不想引起混淆。
一旦我们以这种方式定义了(反)序列化行为,我们的脚本现在可以成功运行:
$ python ../export_attr.py
testing
定义接受或返回绑定 C++类的自定义运算符
一旦您定义了一个自定义 C++类,您也可以将该类用作自定义运算符(即自由函数)的参数或返回值。假设您有一个以下自由函数:
c10::intrusive_ptr<MyStackClass<std::string>> manipulate_instance(const c10::intrusive_ptr<MyStackClass<std::string>>& instance) {
instance->pop();
return instance;
}
您可以在您的 TORCH_LIBRARY
块中运行以下代码进行注册:
m.def(
"manipulate_instance(__torch__.torch.classes.my_classes.MyStackClass x) -> __torch__.torch.classes.my_classes.MyStackClass Y",
manipulate_instance
);
更多关于注册 API 的详细信息,请参考自定义操作教程。
完成此操作后,您可以使用如下示例中的操作:
class TryCustomOp(torch.nn.Module):
def __init__(self):
super(TryCustomOp, self).__init__()
self.f = torch.classes.my_classes.MyStackClass(["foo", "bar"])
def forward(self):
return torch.ops.my_classes.manipulate_instance(self.f)
备注
将一个以 C++类作为参数的操作符进行注册,需要确保自定义类已经注册。您可以通过确保自定义类的注册和您的自由函数定义在同一 TORCH_LIBRARY
块中,并且自定义类注册在前面来实现这一点。将来,我们可能会放宽这一要求,使得这些可以在任何顺序中注册。
结论 ¶
本教程向您介绍了如何将 C++ 类暴露给 TorchScript(以及 Python),如何注册其方法,如何从 Python 和 TorchScript 使用该类,以及如何使用该类保存和加载代码,并在独立的 C++ 进程中运行该代码。现在,您已经准备好扩展您的 TorchScript 模型,使用 C++ 类与第三方 C++ 库接口或实现任何其他需要 Python、TorchScript 和 C++ 之间无缝融合的用例。
如往常一样,如果您遇到任何问题或有疑问,您可以使用我们的论坛或 GitHub 问题来联系。此外,我们的常见问题解答(FAQ)页面可能包含有用的信息。