使用自定义C++类扩展TorchScript
创建于:2020年1月23日 | 最后更新:2024年12月2日 | 最后验证:2024年11月5日
警告
TorchScript 不再处于积极开发阶段。
本教程是自定义操作符教程的后续,介绍了我们为将C++类同时绑定到TorchScript和Python而构建的API。该API与pybind11非常相似,如果您熟悉该系统,大多数概念都可以直接迁移过来。
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
是您需要包含的头文件,以便使用您的自定义类扩展 TorchScript。请注意,每当我们处理自定义类的实例时,我们都是通过
c10::intrusive_ptr<>
的实例来进行的。可以将intrusive_ptr
视为类似于std::shared_ptr
的智能指针,但引用计数直接存储在对象中,而不是存储在单独的元数据块中(如std::shared_ptr
所做的那样)。torch::Tensor
内部使用相同的指针类型;自定义类也必须使用这种指针类型,以便我们能够一致地管理不同的对象类型。第二件需要注意的事情是,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。
现在让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将示例构建为C++项目
现在,我们将使用CMake构建系统来构建上述C++代码。首先,将我们到目前为止涵盖的所有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
文件并将其放在同一目录中。以下是CMakeLists.txt
应该看起来的样子:
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。您的文件树应如下所示:
custom_class_project/
class.cpp
CMakeLists.txt
build/
我们假设您已经按照上一个教程中描述的方式设置了您的环境。继续并调用cmake,然后使用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
你会发现现在(除了其他东西之外)在构建目录中有一个动态库文件。在Linux上,这可能被命名为libcustom_class.so
。所以文件树应该看起来像:
custom_class_project/
class.cpp
CMakeLists.txt
build/
libcustom_class.so
从Python和TorchScript使用C++类
现在我们已经将我们的类及其注册编译成一个.so
文件,
我们可以将该.so加载到Python中并尝试使用它。以下是一个演示脚本:
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
我们也可以在C++进程中使用自定义注册的C++类,通过libtorch。例如,让我们定义一个简单的nn.Module
,它实例化并调用我们的MyStackClass类上的一个方法:
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')
foo.pt
在我们的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。有关如何执行此操作的完整说明,请参阅Loading a TorchScript Model in C++ Tutorial。
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,以及来自上面的 custom_class_project
的源代码树。我们将把 custom_class_project
作为依赖项添加到这个 C++ 项目中,以便我们可以将自定义类构建到二进制文件中。
让我们用以下内容填充 infer.cpp
:
#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;
}
同样地,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)
你知道流程: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令人兴奋的C++二进制文件了:
$ ./infer
momfoobarbaz
难以置信!
将自定义类移入/移出 IValues
你可能还需要将自定义类移入或移出IValue``s, 例如 当你从 TorchScript 方法中 获取或返回 ``IValue``s 时,
或者 你想在 C++ 中实例化一个自定义类属性。 为了从自定义的 C++ 类实例创建
``IValue
:
torch::make_custom_class
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,调用与这些参数匹配的 T 的构造函数,并将该实例包装并返回。 然而,它不仅仅返回一个指向自定义类对象的指针,而是返回一个包装该对象的 IValue
。然后你可以直接将这个IValue
传递给 TorchScript。如果你已经有一个指向你的类的
intrusive_ptr
,你可以直接使用构造函数IValue(intrusive_ptr
从中构造一个IValue。)
为了将IValue
转换回自定义类:
IValue::toCustomClass
将返回一个指向() IValue
包含的自定义类的intrusive_ptr
。在内部,此函数会检查T
是否注册为自定义类,并且IValue
确实包含一个自定义类。你可以通过调用isCustomClass()
手动检查IValue
是否包含自定义类。
为自定义C++类定义序列化/反序列化方法
如果你尝试保存一个带有自定义绑定的C++类作为属性的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++类中保存哪些信息。您必须手动指定。方法是使用class_
上的特殊def_pickle
方法在类上定义__getstate__
和__setstate__
方法。
注意
TorchScript 中 __getstate__
和 __setstate__
的语义与 Python pickle 模块的语义相同。您可以
阅读更多
关于我们如何使用这些方法的信息。
这里是一个我们可以添加到MyStackClass
注册中的def_pickle
调用示例,以包含序列化方法:
// 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));
});
注意
我们在pickle API中采取了与pybind11不同的方法。pybind11使用一个特殊的函数pybind11::pickle()
,你可以将其传递给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,我们的脚本现在就可以成功运行了:
$ python ../export_attr.py
testing
定义接受或返回绑定C++类的自定义运算符
一旦你定义了一个自定义的C++类,你也可以将该类作为参数或从自定义操作符(即自由函数)返回。假设你有以下自由函数:
c10::intrusive_ptr<MyStackClass<std::string>> manipulate_instance(const c10::intrusive_ptr<MyStackClass<std::string>>& instance) {
instance->pop();
return instance;
}
你可以在你的TORCH_LIBRARY
块中运行以下代码来注册它:
m.def(
"manipulate_instance(__torch__.torch.classes.my_classes.MyStackClass x) -> __torch__.torch.classes.my_classes.MyStackClass Y",
manipulate_instance
);
有关注册API的更多详细信息,请参阅自定义操作教程。
完成此操作后,您可以像以下示例一样使用该操作:
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)
注意
注册一个以C++类为参数的运算符要求自定义类已经被注册。你可以通过确保自定义类注册和你的自由函数定义在同一个TORCH_LIBRARY
块中,并且自定义类注册先于自由函数定义来强制执行这一点。将来,我们可能会放宽这一要求,以便这些可以以任何顺序注册。
结论
本教程向您介绍了如何将C++类暴露给TorchScript(以及扩展的Python),如何注册其方法,如何从Python和TorchScript中使用该类,以及如何保存和加载使用该类的代码,并在独立的C++进程中运行该代码。您现在可以准备使用与第三方C++库接口的C++类来扩展您的TorchScript模型,或实现任何其他需要Python、TorchScript和C++之间无缝融合的用例。
一如既往,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 来联系我们。此外,我们的 常见问题解答 (FAQ) 页面 可能包含有用的信息。