使用自定义 C++ 类扩展 TorchScript¶
Created On: Jan 23, 2020 | Last Updated: Dec 02, 2024 | Last Verified: Nov 05, 2024
警告
TorchScript 不再处于活跃开发状态。
本教程是 自定义操作符 教程的后续内容,介绍了我们构建的 API,用于同时将 C++ 类绑定到 TorchScript 和 Python。这个 API 非常类似于 pybind11,如果你熟悉该系统,许多概念都可以直接迁移过来。
在 C++ 中实现并绑定类¶
在本教程中,我们将定义一个简单的 C++ 类,该类在成员变量中保持持久状态。
// 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);
}
}
};
需要注意以下几点:
torch/custom_class.h
是扩展 TorchScript 所需包含的头文件。注意,每当我们使用自定义类的实例时,我们都是通过
c10::intrusive_ptr<>
实例来操作。可将intrusive_ptr
类比作std::shared_ptr
的智能指针,但它的引用计数直接存储在对象中,而不是像std::shared_ptr
那样存储在一个单独的元数据块中。torch::Tensor
内部使用同样的指针类型;而自定义类也必须使用此指针类型,以便我们能够一致地管理不同对象类型。第二点需要注意的是用户定义的类必须继承自
torch::CustomClassHolder
。这样可以确保自定义类有空间存储引用计数。
现在让我们看看如何使此类对 TorchScript 可见,这一过程称为 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,将迄今为止介绍的所有 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,并将其放在相同目录下。以下是 CMakeLists.txt
的内容:
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。你的文件树应该像这样:
custom_class_project/
class.cpp
CMakeLists.txt
build/
我们假设你按照 之前的教程 设置了你的环境。继续调用 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(以及其他文件)在 build 目录下出现了一个动态库文件。在 Linux 上,这个文件很可能名为 libcustom_class.so
。因此文件树应该看起来像这样:
custom_class_project/
class.cpp
CMakeLists.txt
build/
libcustom_class.so
在 Python 和 TorchScript 中使用 C++ 类¶
现在我们已经将类及其注册编译成一个 .so
文件,我们可以将此 .so 文件加载到 Python 中并试验它。以下是演示代码:
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¶
我们还可以在使用 libtorch 的 C++ 进程中使用自定义注册的 C++类。作为示例,让我们定义一个简单的``nn.Module``,它实例化并调用我们 MyStackClass 类中的方法:
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')
现在我们的文件系统中已经包含了我们刚刚定义的序列化 TorchScript 程序 foo.pt
。
现在,我们将定义一个新的 CMake 项目,以展示如何加载这个模型及其所需的 .so 文件。关于如何全面实现这一点,请参考 在 C++ 中加载 TorchScript 模型的教程。
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,以及上述 custom_class_project
的源代码树。我们将添加 custom_class_project
作为此C++ 项目的依赖项,以便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;
}
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)
按照流程: cd build
,cmake
,然后运行 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
现在我们可以运行令人激动的 C++ 二进制文件了:
$ ./infer
momfoobarbaz
不可思议!
将自定义类移动到/从 IValues 中¶
你可能还需要将自定义类移动到或移出 IValue
中,例如当你从 TorchScript 方法中获取或返回 IValue
,或者希望在 C++ 中实例化一个自定义类属性时。对于从自定义 C++ 类实例创建 IValue
:
torch::make_custom_class<T>()
提供了一个类似于 c10::intrusive_ptr<T> 的 API,它会接收你提供给它的任意参数集,调用匹配该参数集的 T 的构造函数,并包装该实例后返回。然而,它不是仅返回指向自定义类对象的指针,而是返回一个带有对象包装的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类的
intrusive_ptr
,你可以使用构造函数IValue(intrusive_ptr<T>)
直接从它构造一个 IValue。
对于将 IValue
转换回自定义类:
IValue::toCustomClass<T>()
会返回一个intrusive_ptr<T>
,指向IValue
包含的自定义类。内部,这个函数会检查T
是否已经注册为自定义类,并检查IValue
是否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以自定义绑定 C++ 类为属性的 ScriptModule
,你会收到如下错误消息:
# 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 无法自动确定要从你的 C++ 类中保存哪些信息。你必须手动指定。实现方式是为类定义 __getstate__
和 __setstate__
方法,使用 class_
的特殊 def_pickle
方法。
备注
TorchScript 中 __getstate__
和 __setstate__
的语义与 Python pickle 模块的语义相同。关于我们如何使用这些方法,可以 阅读更多。
以下是我们可以添加到 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));
});
备注
我们在 pickle API 上采用了不同于 pybind11 的方法。pybind11 提供了一个名为 pybind11::pickle()
的特殊函数,你可以将其传递给 class_::def()
;而我们为了这个目的有一个单独的方法 def_pickle
。这是因为``torch::jit::pickle`` 这个名称已经被占用了,我们不希望引起混乱。
一旦以这种方式定义了(反)序列化行为,我们的脚本现在可以成功运行:
$ python ../export_attr.py
testing
定义采用或返回绑定 C++ 类的自定义操作符¶
一旦你定义了一个自定义 C++ 类,你也可以使用该类作为参数或返回值来自定义操作符(即自由函数)。假设你有如下自由函数:
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
块内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
);
有关注册 API 的详细信息,请参考 自定义操作符教程。
完成后,你可以像以下示例一样使用该操作符:
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
注册一个以 C++ 类为参数的操作符要求该自定义类已被注册。你可以通过确保自定义类注册与你的自由函数定义在同一个 TORCH_LIBRARY
块中,并且自定义类注册优先来强制执行这一点。在未来,我们可能会放松这个要求,以允许这些可以按任何顺序注册。
总结¶
本教程向你介绍了如何将 C++ 类公开给 TorchScript(以及扩展到 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。你现在已经准备好通过接口第三方 C++库或实现任何其他需要在 Python、TorchScript 和 C++之间平滑交互的用例,使用 C++ 类扩展你的 TorchScript 模型了。
一如既往,如果您遇到任何问题或有任何疑问,可以使用我们的 论坛 或 GitHub问题 联系我们。此外,我们的 常见问题解答 (FAQ) 页 可能也有有用的信息。