使用自定义 C ++类扩展 TorchScript
原文: https://pytorch.org/tutorials/advanced/torch_script_custom_classes.html
本教程是自定义运算符教程的后续教程,并介绍了我们为将 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,则大多数概念都将转移过来。
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 Stack : torch::jit::CustomClassHolder {
std::vector<T> stack_;
Stack(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<Stack> clone() const {
return c10::make_intrusive<Stack>(stack_);
}
void merge(const c10::intrusive_ptr<Stack>& c) {
for (auto& elem : c->stack_) {
push(elem);
}
}
};
有几件事要注意:
torch/custom_class.h
是您需要使用自定义类扩展 TorchScript 的标头。- 注意,无论何时使用自定义类的实例,我们都通过
c10::intrusive_ptr<>
的实例来实现。 将intrusive_ptr
视为类似于std::shared_ptr
的智能指针。 使用此智能指针的原因是为了确保在语言(C ++,Python 和 TorchScript)之间对对象实例进行一致的生命周期管理。 - 注意的第二件事是用户定义的类必须继承自
torch::jit::CustomClassHolder
。 这确保了所有设置都可以处理前面提到的生命周期管理系统。
现在让我们看一下如何使该类对 TorchScript 可见,该过程称为_绑定_该类:
// Notice a few things:
// - We pass the class to be registered as a template parameter to
// `torch::jit::class_`. In this instance, we've passed the
// specialization of the Stack class ``Stack<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 single parameter to ``torch::jit::class_()`` is a
// string indicating the name of the class. This is the name
// the class will appear as in both Python and TorchScript.
// For example, our Stack class would appear as ``torch.classes.Stack``.
static auto testStack =
torch::jit::class_<Stack<std::string>>("Stack")
// The following line registers the contructor of our Stack
// class that takes a single `std::vector<std::string>` argument,
// i.e. it exposes the C++ method `Stack(std::vector<T> init)`.
// Currently, we do not support registering overloaded
// constructors, so for now you can only `def()` one instance of
// `torch::jit::init`.
.def(torch::jit::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<Stack<std::string>>& self) {
return self->stack_.back();
})
// The following four lines expose methods of the Stack<std::string>
// class as-is. `torch::jit::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", &Stack<std::string>::push)
.def("pop", &Stack<std::string>::pop)
.def("clone", &Stack<std::string>::clone)
.def("merge", &Stack<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,将到目前为止介绍的所有 C ++代码放入class.cpp
文件中。 然后,编写一个简单的CMakeLists.txt
文件并将其放置在同一目录中。 CMakeLists.txt
的外观如下:
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}")
另外,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/
现在,要构建项目,请继续从 PyTorch 网站下载适当的 libtorch 二进制文件。 将 zip 存档解压缩到某个位置(在项目目录中可能很方便),并记下将其解压缩到的路径。 接下来,继续调用 cmake,然后进行构建项目:
$ cd build
$ cmake -DCMAKE_PREFIX_PATH=/path/to/libtorch ..
-- 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
您会发现,构建目录中现在有一个动态库文件。 在 Linux 上,它可能名为libcustom_class.so
。 因此,文件树应如下所示:
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为.so
文件,我们可以将 <cite>.so</cite> 加载到 Python 中并进行尝试。 这是一个演示脚本的脚本:
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("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 Stack(std::vector<T> init) constructor
# we registered earlier
s = torch.classes.Stack(["foo", "bar"])
# We can call methods in Python
s.push("pushed")
assert s.pop() == "pushed"
# 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.
Stack = torch.classes.Stack
@torch.jit.script
def do_stacks(s : Stack): # We can pass a custom class instance to TorchScript
s2 = torch.classes.Stack(["hi", "mom"]) # We can instantiate the class
s2.merge(s) # We can call a method on the class
return s2.clone(), s2.top() # We can also return instances of the class
# from TorchScript function/methods
stack, top = do_stacks(torch.classes.Stack(["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
我们也可以在使用 libtorch 的 C ++进程中使用自定义注册的 C ++类。 举例来说,让我们定义一个简单的nn.Module
,该实例在我们的 Stack 类上实例化并调用一个方法:
import torch
torch.classes.load_library('libcustom_class.so')
class Foo(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, s : str) -> str:
stack = torch.classes.Stack(["hi", "mom"])
return stack.pop() + s
scripted_foo = torch.jit.script(Foo())
print(scripted_foo.graph)
scripted_foo.save('foo.pt')
我们文件系统中的foo.pt
现在包含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,请查看在 C ++教程中加载 TorchScript 模型。
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/
请注意,我们已经复制了序列化的foo.pt
文件以及上面custom_class_project
的源代码树。 我们将添加custom_class_project
作为对此 C ++项目的依赖项,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::script::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,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)
您知道练习:cd build
,cmake
和make
:
$ cd build
$ cmake -DCMAKE_PREFIX_PATH=/path/to/libtorch ..
-- 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兴奋的 C ++二进制文件:
$ ./infer
momfoobarbaz
难以置信!
定义自定义 C ++类的序列化/反序列化方法
如果您尝试将具有自定义绑定 C ++类的ScriptModule
保存为属性,则会出现以下错误:
# export_attr.py
import torch
torch.classes.load_library('libcustom_class.so')
class Foo(torch.nn.Module):
def __init__(self):
super().__init__()
self.stack = torch.classes.Stack(["just", "testing"])
def forward(self, s : str) -> str:
return self.stack.pop() + s
scripted_foo = torch.jit.script(Foo())
scripted_foo.save('foo.pt')
$ python export_attr.py
RuntimeError: Cannot serialize custom bound C++ class __torch__.torch.classes.Stack. Please define serialization methods via torch::jit::pickle_ for this class. (pushIValueImpl at ../torch/csrc/jit/pickler.cpp:128)
这是因为 TorchScript 无法自动找出 C ++类中保存的信息。 您必须手动指定。 这样做的方法是使用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。 您可以阅读更多有关如何使用这些方法的信息。
这是一个如何更新Stack
类的注册码以包含序列化方法的示例:
static auto testStack =
torch::jit::class_<Stack<std::string>>("Stack")
.def(torch::jit::init<std::vector<std::string>>())
.def("top", [](const c10::intrusive_ptr<Stack<std::string>>& self) {
return self->stack_.back();
})
.def("push", &Stack<std::string>::push)
.def("pop", &Stack<std::string>::pop)
.def("clone", &Stack<std::string>::clone)
.def("merge", &Stack<std::string>::merge)
// 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<Stack<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<Stack<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 Stack<std::string>
// and call the single-argument std::vector<std::string>
// constructor with the serialized state.
return c10::make_intrusive<Stack<std::string>>(std::move(state));
});
Note
我们采用与 pickle API 中的 pybind11 不同的方法。 pybind11 作为传递给class_::def()
的特殊功能pybind11::pickle()
,为此我们有一个单独的方法def_pickle
。 这是因为名称torch::jit::pickle
已经被使用,我们不想引起混淆。
以这种方式定义(反)序列化行为后,脚本现在可以成功运行:
import torch
torch.classes.load_library('libcustom_class.so')
class Foo(torch.nn.Module):
def __init__(self):
super().__init__()
self.stack = torch.classes.Stack(["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
testing
结论
本教程向您介绍了如何向 TorchScript(以及扩展为 Python)公开 C ++类,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以及运行该代码。 在独立的 C ++过程中。 现在,您可以使用与第三方 C ++库接口的 C ++类扩展 TorchScript 模型,或实现需要 Python,TorchScript 和 C ++之间的界线才能平滑融合的任何其他用例。
与往常一样,如果您遇到任何问题或疑问,可以使用我们的论坛或 GitHub 问题进行联系。 另外,我们的常见问题解答(FAQ)页面可能包含有用的信息。