Skip to content

GitLab

Commit b2735663 authored by liucong8560's avatar liucong8560
Browse files

Merge branch 'develop' into 'master' 

Develop

See merge request !2
parents 5a36110c f6ecdce1
Hide whitespace changes
Inline Side-by-side
Showing with 151 additions and 557 deletions
3rdParty/InstallOpenCVDependences.sh deleted 100644 → 0
View file @ 5a36110c
#! /bin/sh
############### Ubuntu ###############
# 参考：https://docs.opencv.org/3.4.11/d7/d9f/tutorial_linux_install.html
# apt-get install build-essential -y
# apt-get install cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev -y
# apt-get install python-dev python-numpy libtbb2 libtbb-dev libjpeg-dev libpng-dev libtiff-dev libjasper-dev libdc1394-22-dev -y # 处理图像所需的包，可选
############### CentOS ###############
yum install gcc gcc-c++ gtk2-devel gimp-devel gimp-devel-tools gimp-help-browser zlib-devel libtiff-devel libjpeg-devel libpng-devel gstreamer-devel libavc1394-devel libraw1394-devel libdc1394-devel jasper-devel jasper-utils swig python libtool nasm -y
\ No newline at end of file
CMakeLists.txt
View file @ b2735663
......@@ -2,7 +2,7 @@
cmake_minimum_required(VERSION 3.5)
# 设置项目名
project(MIGraphX_Samples)
project(GPT2)
# 设置编译器
set(CMAKE_CXX_COMPILER g++)
......@@ -12,7 +12,6 @@ set(CMAKE_BUILD_TYPE release)
# 添加头文件路径
set(INCLUDE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/Src/
${CMAKE_CURRENT_SOURCE_DIR}/Src/Utility/
${CMAKE_CURRENT_SOURCE_DIR}/Src/NLP/GPT2/
$ENV{DTKROOT}/include/
${CMAKE_CURRENT_SOURCE_DIR}/depend/include/)
include_directories(${INCLUDE_PATH})
......@@ -23,11 +22,7 @@ set(LIBRARY_PATH ${CMAKE_CURRENT_SOURCE_DIR}/depend/lib64/
link_directories(${LIBRARY_PATH})
# 添加依赖库
set(LIBRARY opencv_core
opencv_imgproc
opencv_imgcodecs
opencv_dnn
migraphx_ref
set(LIBRARY migraphx_ref
migraphx
migraphx_c
migraphx_device
......@@ -37,12 +32,10 @@ link_libraries(${LIBRARY})
# 添加源文件
set(SOURCE_FILES ${CMAKE_CURRENT_SOURCE_DIR}/Src/main.cpp
${CMAKE_CURRENT_SOURCE_DIR}/Src/Sample.cpp
${CMAKE_CURRENT_SOURCE_DIR}/Src/NLP/GPT2/tokenization.cpp
${CMAKE_CURRENT_SOURCE_DIR}/Src/NLP/GPT2/utf8proc.c
${CMAKE_CURRENT_SOURCE_DIR}/Src/NLP/GPT2/GPT2.cpp
${CMAKE_CURRENT_SOURCE_DIR}/Src/Utility/CommonUtility.cpp
${CMAKE_CURRENT_SOURCE_DIR}/Src/GPT2.cpp
${CMAKE_CURRENT_SOURCE_DIR}/Src/Utility/tokenization.cpp
${CMAKE_CURRENT_SOURCE_DIR}/Src/Utility/utf8proc.c
${CMAKE_CURRENT_SOURCE_DIR}/Src/Utility/Filesystem.cpp)
# 添加可执行目标
add_executable(MIGraphX_Samples ${SOURCE_FILES})
add_executable(GPT2 ${SOURCE_FILES})
Doc/Images/GPT_04.png
View replaced file @ 5a36110c View file @ b2735663
Doc/Images/GPT_04.png

35.3 KB | W: | H:

Doc/Images/GPT_04.png

14.1 KB | W: | H:

Doc/Images/GPT_04.png
Doc/Images/GPT_04.png
Doc/Images/GPT_04.png
Doc/Images/GPT_04.png
  • 2-up
  • Swipe
  • Onion skin
Doc/Tutorial_Cpp/GPT2.md → Doc/Tutorial_Cpp.md
View file @ b2735663
......@@ -12,9 +12,9 @@ GPT（Generative Pre-trained Transformer）系列模型以不断堆叠transforme
| GPT-2 | 2019 年 2 月 | 15 亿 | 40GB |
| GPT-3 | 2020 年 5 月 | 1,750 亿 | 45TB |
本次采用GPT-2模型进行诗词生成任务，模型文件下载链接：https://pan.baidu.com/s/1KWeoUuakCZ5dualK69qCcw, 提取码：4pmh。将GPT2_shici.onnx模型文件保存在Resource/Models/NLP/GPT2文件夹下。整体模型结构如下图所示，也可以通过netron工具：https://netron.app/ 查看GPT-2的模型结构。
本次采用GPT-2模型进行诗词生成任务，模型文件下载链接：https://pan.baidu.com/s/1KWeoUuakCZ5dualK69qCcw, 提取码：4pmh。将GPT2_shici.onnx模型文件保存在Resource/文件夹下。整体模型结构如下图所示，也可以通过netron工具：https://netron.app/ 查看GPT-2的模型结构。
<img src="../Images/GPT_01.png" style="zoom:100%;" align=middle>
<img src="./Images/GPT_01.png" style="zoom:100%;" align=middle>
## 预处理
......@@ -27,11 +27,11 @@ GPT（Generative Pre-trained Transformer）系列模型以不断堆叠transforme
首先，根据提供的词汇表路径，通过cuBERT::FullTokenizer()函数加载词汇表，用于后续对输入文本的编码操作。其次，将词汇表中的内容依次保存到vector容器output中，用于数据后处理中的解码操作。
```c++
cuBERT::FullTokenizer tokenizer = cuBERT::FullTokenizer("../Resource/Models/NLP/GPT2/vocab_shici.txt");
cuBERT::FullTokenizer tokenizer = cuBERT::FullTokenizer("../Resource/vocab_shici.txt");
std::ifstream infile;
std::string buf;
std::vector<std::string> output;
infile.open("../Resource/Models/NLP/GPT2/vocab_shici.txt");
infile.open("../Resource/vocab_shici.txt");
while (std::getline(infile,buf))
{
output.push_back(buf);
......@@ -46,7 +46,7 @@ ErrorCode GPT2::Preprocessing(cuBERT::FullTokenizer tokenizer,
std::vector<long unsigned int> &input_id)
{
// 对问题进行分词操作
int max_seq_length =1024;
int max_seq_length =1000;
std::vector<std::string> tokens_question;
tokens_question.reserve(max_seq_length);
tokenizer.tokenize(question, &tokens_question, max_seq_length);
......@@ -65,7 +65,7 @@ ErrorCode GPT2::Preprocessing(cuBERT::FullTokenizer tokenizer,
对于GPT-2这种生成式语言模型来说，模型不是仅执行一次推理就结束，而是需要执行多次推理，才能得到最终的答案。如下图所示，GPT-2模型每次推理仅生成一个词，通过将生成的词与输入数据拼接，输入到模型中继续下一次的推理，直到循环结果或者生成[SEP]结束标识符才结束推理。
<img src="../Images/GPT_02.png" style="zoom:70%;" align=middle>
<img src="./Images/GPT_02.png" style="zoom:70%;" align=middle>
具体GPT-2模型的推理，如下代码所示。首先，通过gpt2.Inference()函数实现模型的具体推理细节，推理结果保存在outputs中。其次，对每次推理结果进行判断，当判断为[SEP]结束标志符时，结束循环完成推理，否则就将推理结果outputs加入到输入数据input_id中，继续下一次的模型推理。
......@@ -97,12 +97,12 @@ long unsigned int GPT2::Inference(const std::vector<long unsigned int> &input_id
input[0][j] = input_id[j];
}
// 设置输入shape
// 输入shape
std::vector<std::vector<std::size_t>> inputShapes;
inputShapes.push_back({1,input_id.size()});
// 输入数据
migraphx::parameter_map inputData;
std::unordered_map<std::string, migraphx::argument> inputData;
inputData[inputName]=migraphx::argument{migraphx::shape(inputShape.type(),inputShapes[0]),(long unsigned int*)input};
// 推理
......
Doc/Tutorial_Python/GPT2.md → Doc/Tutorial_Python.md
View file @ b2735663
......@@ -12,9 +12,9 @@ GPT（Generative Pre-trained Transformer）系列模型以不断堆叠的transfo
| GPT-2 | 2019 年 2 月 | 15 亿 | 40GB |
| GPT-3 | 2020 年 5 月 | 1,750 亿 | 45TB |
本次采用GPT-2模型进行诗词生成任务，模型文件下载链接：https://pan.baidu.com/s/1KWeoUuakCZ5dualK69qCcw , 提取码：4pmh 。将GPT2_shici.onnx模型文件保存在Resource/Models/NLP/GPT2文件夹下。整体模型结构如下图所示，也可以通过netron工具：https://netron.app/ 查看GPT-2的模型结构。
本次采用GPT-2模型进行诗词生成任务，模型文件下载链接：https://pan.baidu.com/s/1KWeoUuakCZ5dualK69qCcw , 提取码：4pmh 。将GPT2_shici.onnx模型文件保存在Resource/文件夹下。整体模型结构如下图所示，也可以通过netron工具：https://netron.app/ 查看GPT-2的模型结构。
<img src="../Images/GPT_01.png" style="zoom:100%;" align=middle>
<img src="./Images/GPT_01.png" style="zoom:100%;" align=middle>
## 预处理
......@@ -28,7 +28,7 @@ GPT（Generative Pre-trained Transformer）系列模型以不断堆叠的transfo
```python
# 加载词汇表
vocab_file = os.path.join('../../../Resource/Models/NLP/GPT2', 'vocab_shici.txt')
vocab_file = os.path.join('../Resource/', 'vocab_shici.txt')
tokenizer = BertTokenizerFast(vocab_file, sep_token="[SEP]", pad_token="[PAD]", cls_token="[CLS]")
```
......@@ -63,7 +63,7 @@ maxInput={"input":[1,1000]}
for _ in range(max_len):
# 推理
result = model.run({inputName: migraphx.argument(input_ids)})
result = model.run({inputName: input_ids})
logits = [float(x) for x in result[0].tolist()]
...
......@@ -83,7 +83,7 @@ for _ in range(max_len):
2.循环推理，GPT-2模型不像其他模型一样只需要执行一次推理，而是需要循环执行多次推理才能完成。首先，模型推理限定在for循环中，将输入数据input_ids，输入到model.run({...})中执行推理，生成一个token的id。其次，将推理结果拼接到输入数据input_ids中，执行下一次循环。最后，当循环结束或者生成的词为[SEP]结束标志符时，完成GPT-2模型的整体推理。如下图所示，为GPT-2模型的一次完整推理过程。
<img src="../Images/GPT_02.png" style="zoom:70%;" align=middle>
<img src="./Images/GPT_02.png" style="zoom:70%;" align=middle>
## 数据后处理
......
Python/NLP/GPT2/gpt2.py → Python/gpt2.py
View file @ b2735663
......@@ -5,7 +5,7 @@ import migraphx
# 加载词汇表
print("INFO: Complete loading the vocabulary")
vocab_file = os.path.join('../../../Resource/Models/NLP/GPT2', 'vocab_shici.txt')
vocab_file = os.path.join('../Resource/', 'vocab_shici.txt')
tokenizer = BertTokenizerFast(vocab_file, sep_token="[SEP]", pad_token="[PAD]", cls_token="[CLS]")
# 设置最大输入shape
......@@ -13,7 +13,7 @@ maxInput={"input":[1,1000]}
# 加载模型
print("INFO: Parsing and compiling the model")
model = migraphx.parse_onnx("../../../Resource/Models/NLP/GPT2/GPT2_shici.onnx", map_input_dims=maxInput)
model = migraphx.parse_onnx("../Resource/GPT2_shici.onnx", map_input_dims=maxInput)
inputName=model.get_parameter_names()[0]
inputShape=model.get_parameter_shapes()[inputName].lens()
print("inputName:{0} \ninputShape:{1}".format(inputName,inputShape))
......
README.md
View file @ b2735663
......@@ -6,30 +6,72 @@ GPT2模型：第二代生成式预训练模型（Generative Pre-Training2）。
## 模型结构
GPT2主要使用Transformer的Decoder模块为特征提取器，并对Transformer Decoder进行了一些改动，原本的Decoder包含了两个Multi-Head Attention结构，而GPT2只保留了Mask Multi-Head Attention。
## 构建安装
## Python版本推理
在光源可拉取推理的docker镜像，GPT2模型推理的镜像如下：
本次采用GPT-2模型进行诗词生成任务，模型文件下载链接：https://pan.baidu.com/s/1KWeoUuakCZ5dualK69qCcw , 提取码：4pmh ，并将GPT2_shici.onnx模型文件保存在Resource/文件夹下。下面介绍如何运行python代码示例，Python示例的详细说明见Doc目录下的Tutorial_Python.md。
### 下载镜像
在光源中下载MIGraphX镜像：
```python
docker pull image.sourcefind.cn:5000/dcu/admin/base/custom:ort1.14.0_migraphx3.0.0-dtk22.10.1
```
### 安装Opencv依赖
### 设置Python环境变量
```
export PYTHONPATH=/opt/dtk/lib:$PYTHONPATH
```
### 安装依赖
```python
cd <path_to_migraphx_samples>
sh ./3rdParty/InstallOpenCVDependences.sh
# 进入gpt2 migraphx工程根目录
cd <path_to_gpt2_migraphx>
# 进入示例程序目录
cd ./Python/
# 安装依赖
pip install -r requirements.txt
```
### 修改CMakeLists.txt
### 设置动态shape模式
```python
export MIGRAPHX_DYNAMIC_SHAPE=1
```
### 运行示例
在Python目录下执行如下命令运行该示例程序：
```python
python gpt2.py
```
如下所示，采用交互式界面，通过输入开头诗词，GPT2模型可以生成后续的诗句。
<img src="./Doc/Images/GPT_03.png" style="zoom:80%;" align=middle>
## C++版本推理
本次采用GPT-2模型进行诗词生成任务，模型文件下载链接：https://pan.baidu.com/s/1KWeoUuakCZ5dualK69qCcw , 提取码：4pmh ，并将GPT2_shici.onnx模型文件保存在Resource/文件夹下。下面介绍如何运行C++代码示例，C++示例的详细说明见Doc目录下的Tutorial_Cpp.md。
### 下载镜像
```
docker pull image.sourcefind.cn:5000/dcu/admin/base/custom:ort1.14.0_migraphx3.0.0-dtk22.10.1
```
- 如果使用ubuntu系统，需要修改CMakeLists.txt中依赖库路径：
将"${CMAKE_CURRENT_SOURCE_DIR}/depend/lib64/"修改为"${CMAKE_CURRENT_SOURCE_DIR}/depend/lib/"
### 修改CMakeLists.txt
- **MIGraphX2.3.0及以上版本需要c++17**
如果使用ubuntu系统，需要修改CMakeLists.txt中依赖库路径：
将"${CMAKE_CURRENT_SOURCE_DIR}/depend/lib64/"修改为"${CMAKE_CURRENT_SOURCE_DIR}/depend/lib/"
### 安装OpenCV并构建工程
### 构建工程
```
rbuild build -d depend
......@@ -42,13 +84,13 @@ rbuild build -d depend
**Centos**:
```
export LD_LIBRARY_PATH=<path_to_migraphx_samples>/depend/lib64/:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=<path_to_gpt2_migraphx>/depend/lib64/:$LD_LIBRARY_PATH
```
**Ubuntu**:
```
export LD_LIBRARY_PATH=<path_to_migraphx_samples>/depend/lib/:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=<path_to_gpt2_migraphx>/depend/lib/:$LD_LIBRARY_PATH
```
然后执行:
......@@ -57,61 +99,23 @@ export LD_LIBRARY_PATH=<path_to_migraphx_samples>/depend/lib/:$LD_LIBRARY_PATH
source ~/.bashrc
```
## 推理
本次采用GPT-2模型进行诗词生成任务，模型文件下载链接：https://pan.baidu.com/s/1KWeoUuakCZ5dualK69qCcw , 提取码：4pmh ，并将GPT2_shici.onnx模型文件保存在model文件夹下。下面介绍如何运行python代码和C++代码示例，具体推理代码解析，在Doc目录中有详细说明。
### python版本推理
1.参考《MIGraphX教程》中的安装方法安装MIGraphX并设置好PYTHONPATH
2.安装依赖：
```python
# 进入migraphx samples工程根目录
cd <path_to_migraphx_samples>
# 进入示例程序目录
cd Python/NLP/GPT2
### 设置动态shape模式
# 安装依赖
pip install -r requirements.txt
```
3.设置环境变量：
```python
# 设置动态shape模式
export MIGRAPHX_DYNAMIC_SHAPE=1
```
4.在Python/NLP/GPT2目录下执行如下命令运行该示例程序：
### 运行示例
```python
python gpt2.py
```
如下所示，采用交互式界面，通过输入开头诗词，GPT2模型可以生成后续的诗句。
<img src="./Doc/Images/GPT_03.png" style="zoom:80%;" align=middle>
### C++版本推理
切换到build目录中，执行如下命令：
# 进入gpt2 migraphx工程根目录
cd <path_to_gpt2_migraphx>
```python
# 进入build目录
cd ./build/
./MIGraphX_Samples
```
根据提示选择运行GPT2模型的示例程序
```python
# 设置动态shape模式
export MIGRAPHX_DYNAMIC_SHAPE=1
# 运行示例
./MIGraphX_Samples 0
# 执行示例程序
./GPT2
```
如下所示，采用交互式界面，通过输入开头诗词，GPT2模型可以推理出后续的诗句。
......@@ -124,6 +128,4 @@ https://developer.hpccube.com/codes/modelzoo/gpt2_migraphx
## 参考资料
https://github.com/yangjianxin1/GPT2-chitchat
https://github.com/Morizeyao/GPT2-Chinese
\ No newline at end of file
Resource/Configuration.xml deleted 100644 → 0
View file @ 5a36110c
<?xml version="1.0" encoding="GB2312"?>
<opencv_storage>
<!--GPT2-->
<GPT2>
<ModelPath>"../Resource/Models/NLP/GPT2/GPT2_shici.onnx"</ModelPath>
</GPT2>
</opencv_storage>
Src/NLP/GPT2/GPT2.cpp → Src/GPT2.cpp
View file @ b2735663
#include <fstream>
#include <sstream>
#include <GPT2.h>
#include <migraphx/onnx.hpp>
#include <migraphx/gpu/target.hpp>
#include <migraphx/gpu/hip.hpp>
#include <migraphx/generate.hpp>
#include <migraphx/quantization.hpp>
#include <migraphx/reshape2.hpp>
#include <CommonUtility.h>
#include <Filesystem.h>
#include <SimpleLog.h>
#include <algorithm>
#include <string>
#include <vector>
#include <stdexcept>
#include <GPT2.h>
#include <tokenization.h>
namespace migraphxSamples
{
GPT2::GPT2():logFile(NULL)
GPT2::GPT2()
{
}
GPT2::~GPT2()
{
configurationFile.release();
}
ErrorCode GPT2::Initialize(InitializationParameterOfNLP initParamOfNLPGPT2)
ErrorCode GPT2::Initialize()
{
// 初始化(获取日志文件,加载配置文件等)
ErrorCode errorCode=DoCommonInitialization(initParamOfNLPGPT2);
if(errorCode!=SUCCESS)
{
LOG_ERROR(logFile,"fail to DoCommonInitialization\n");
return errorCode;
}
LOG_INFO(logFile,"succeed to DoCommonInitialization\n");
// 获取配置文件参数
FileNode netNode = configurationFile["GPT2"];
std::string modelPath=initializationParameter.parentPath+(std::string)netNode["ModelPath"];
// 获取模型文件
std::string modelPath="../Resource/GPT2_shici.onnx";
// 设置最大输入shape
migraphx::onnx_options onnx_options;
......@@ -51,16 +34,16 @@ ErrorCode GPT2::Initialize(InitializationParameterOfNLP initParamOfNLPGPT2)
// 加载模型
if(Exists(modelPath)==false)
{
LOG_ERROR(logFile,"%s not exist!\n",modelPath.c_str());
LOG_ERROR(stdout,"%s not exist!\n",modelPath.c_str());
return MODEL_NOT_EXIST;
}
net = migraphx::parse_onnx(modelPath, onnx_options);
LOG_INFO(logFile,"succeed to load model: %s\n",GetFileName(modelPath).c_str());
LOG_INFO(stdout,"succeed to load model: %s\n",GetFileName(modelPath).c_str());
// 获取模型输入属性
std::pair<std::string, migraphx::shape> inputAttribute=*(net.get_parameter_shapes().begin());
inputName=inputAttribute.first;
inputShape=inputAttribute.second;
std::unordered_map<std::string, migraphx::shape> inputMap=net.get_parameter_shapes();
inputName=inputMap.begin()->first;
inputShape=inputMap.begin()->second;
// 设置模型为GPU模式
migraphx::target gpuTarget = migraphx::gpu::target{};
......@@ -70,41 +53,8 @@ ErrorCode GPT2::Initialize(InitializationParameterOfNLP initParamOfNLPGPT2)
options.device_id=0; // 设置GPU设备，默认为0号设备
options.offload_copy=true; // 设置offload_copy
net.compile(gpuTarget,options);
LOG_INFO(logFile,"succeed to compile model: %s\n",GetFileName(modelPath).c_str());
return SUCCESS;
}
LOG_INFO(stdout,"succeed to compile model: %s\n",GetFileName(modelPath).c_str());
ErrorCode GPT2::DoCommonInitialization(InitializationParameterOfNLP initParamOfNLPGPT2)
{
initializationParameter = initParamOfNLPGPT2;
// 获取日志文件
logFile=LogManager::GetInstance()->GetLogFile(initializationParameter.logName);
// 加载配置文件
std::string configFilePath=initializationParameter.configFilePath;
if(!Exists(configFilePath))
{
LOG_ERROR(logFile, "no configuration file!\n");
return CONFIG_FILE_NOT_EXIST;
}
if(!configurationFile.open(configFilePath, FileStorage::READ))
{
LOG_ERROR(logFile, "fail to open configuration file\n");
return FAIL_TO_OPEN_CONFIG_FILE;
}
LOG_INFO(logFile, "succeed to open configuration file\n");
// 修改父路径
std::string &parentPath = initializationParameter.parentPath;
if (!parentPath.empty())
{
if(!IsPathSeparator(parentPath[parentPath.size() - 1]))
{
parentPath+=PATH_SEPARATOR;
}
}
return SUCCESS;
}
......@@ -126,8 +76,8 @@ long unsigned int GPT2::Inference(const std::vector<long unsigned int> &input_id
std::vector<std::vector<std::size_t>> inputShapes;
inputShapes.push_back({1,input_id.size()});
// 输入数据
migraphx::parameter_map inputData;
// 创建输入数据
std::unordered_map<std::string, migraphx::argument> inputData;
inputData[inputName]=migraphx::argument{migraphx::shape(inputShape.type(),inputShapes[0]),(long unsigned int*)input};
// 推理
......@@ -163,7 +113,7 @@ ErrorCode GPT2::Preprocessing(cuBERT::FullTokenizer tokenizer,
std::vector<long unsigned int> &input_id)
{
// 分词操作
int max_seq_length =1024;
int max_seq_length =1000;
std::vector<std::string> tokens_question;
tokens_question.reserve(max_seq_length);
tokenizer.tokenize(question, &tokens_question, max_seq_length);
......
Src/NLP/GPT2/GPT2.h → Src/GPT2.h
View file @ b2735663
#ifndef GPT2_H
#define GPT2_H
#ifndef __GPT2_H__
#define __GPT2_H__
#include <cstdint>
#include <string>
#include <migraphx/program.hpp>
#include <CommonDefinition.h>
#include <tokenization.h>
using namespace cuBERT;
namespace migraphxSamples
{
typedef enum _ErrorCode
{
SUCCESS=0,
MODEL_NOT_EXIST,
CONFIG_FILE_NOT_EXIST,
FAIL_TO_LOAD_MODEL,
FAIL_TO_OPEN_CONFIG_FILE,
}ErrorCode;
typedef struct _Predictions
{
long unsigned int index;
......@@ -24,7 +31,7 @@ public:
~GPT2();
ErrorCode Initialize(InitializationParameterOfNLP initParamOfNLPGPT2);
ErrorCode Initialize();
ErrorCode Preprocessing(cuBERT::FullTokenizer tokenizer,
char *question,
......@@ -33,13 +40,6 @@ public:
long unsigned int Inference(const std::vector<long unsigned int> &input_id);
private:
ErrorCode DoCommonInitialization(InitializationParameterOfNLP initParamOfNLPGPT2);
private:
FILE *logFile;
cv::FileStorage configurationFile;
InitializationParameterOfNLP initializationParameter;
migraphx::program net;
std::string inputName;
migraphx::shape inputShape;
......
Src/Sample.cpp deleted 100644 → 0
View file @ 5a36110c
#include <Sample.h>
#include <SimpleLog.h>
#include <GPT2.h>
#include <tokenization.h>
#include <fstream>
using namespace std;
using namespace migraphx;
using namespace migraphxSamples;
void Sample_GPT2()
{
// 加载GPT2模型
GPT2 gpt2;
InitializationParameterOfNLP initParamOfNLPGPT2;
initParamOfNLPGPT2.parentPath = "";
initParamOfNLPGPT2.configFilePath = CONFIG_FILE;
initParamOfNLPGPT2.logName = "";
ErrorCode errorCode = gpt2.Initialize(initParamOfNLPGPT2);
if (errorCode != SUCCESS)
{
LOG_ERROR(stdout, "fail to initialize GPT2!\n");
exit(-1);
}
LOG_INFO(stdout, "succeed to initialize GPT2\n");
// 加载词汇表，用于编码和解码
cuBERT::FullTokenizer tokenizer = cuBERT::FullTokenizer("../Resource/Models/NLP/GPT2/vocab_shici.txt");
std::ifstream infile;
std::string buf;
std::vector<std::string> output;
infile.open("../Resource/Models/NLP/GPT2/vocab_shici.txt");
while (std::getline(infile,buf))
{
output.push_back(buf);
}
std::vector<long unsigned int> input_id;
char question[100];
std::vector<long unsigned int> score;
std::vector<std::string> result;
std::cout << "开始和GPT2对诗，输入CTRL + Z以退出" << std::endl;
while (true)
{
// 数据预处理
std::cout << "question: ";
cin.getline(question, 100);
gpt2.Preprocessing(tokenizer, question, input_id);
// 推理
for(int i=0;i<50;++i)
{
long unsigned int outputs = gpt2.Inference(input_id);
if(outputs == 102)
{
break;
}
input_id.push_back(outputs);
score.push_back(outputs);
}
// 将数值映射为字符
for(int i=0;i<score.size();++i)
{
result.push_back(output[score[i]]);
}
// 打印结果
std::cout << "chatbot: ";
std::cout << question;
for(int j=0; j<result.size();++j)
{
std::cout << result[j];
}
std::cout << std::endl;
// 清除数据
input_id.clear();
result.clear();
score.clear();
}
}
Src/Sample.h deleted 100644 → 0
View file @ 5a36110c
// 示例程序
#ifndef __SAMPLE_H__
#define __SAMPLE_H__
// GPT2 Dynamic sample
void Sample_GPT2();
#endif
\ No newline at end of file
Src/Utility/CommonDefinition.h deleted 100644 → 0
View file @ 5a36110c
// 常用数据类型和宏定义
#ifndef __COMMON_DEFINITION_H__
#define __COMMON_DEFINITION_H__
#include <string>
#include <opencv2/opencv.hpp>
using namespace std;
using namespace cv;
namespace migraphxSamples
{
// 路径分隔符(Linux:‘/’,Windows:’\\’)
#ifdef _WIN32
#define PATH_SEPARATOR '\\'
#else
#define PATH_SEPARATOR '/'
#endif
#define CONFIG_FILE "../Resource/Configuration.xml"
typedef struct __Time
{
string year;
string month;
string day;
string hour;
string minute;
string second;
string millisecond; // ms
string microsecond; // us
string weekDay;
}_Time;
typedef enum _ErrorCode
{
SUCCESS=0, // 0
MODEL_NOT_EXIST, // 模型不存在
CONFIG_FILE_NOT_EXIST, // 配置文件不存在
FAIL_TO_LOAD_MODEL, // 加载模型失败
FAIL_TO_OPEN_CONFIG_FILE, // 加载配置文件失败
IMAGE_ERROR, // 图像错误
}ErrorCode;
typedef struct _ResultOfPrediction
{
float confidence;
int label;
_ResultOfPrediction():confidence(0.0f),label(0){}
}ResultOfPrediction;
typedef struct _ResultOfDetection
{
Rect boundingBox;
float confidence;
int classID;
string className;
bool exist;
_ResultOfDetection():confidence(0.0f),classID(0),exist(true){}
}ResultOfDetection;
typedef struct _InitializationParameterOfDetector
{
std::string parentPath;
std::string configFilePath;
cv::Size inputSize;
std::string logName;
}InitializationParameterOfDetector;
typedef struct _InitializationParameterOfDetector InitializationParameterOfClassifier;
typedef struct _InitializationParameterOfDetector InitializationParameterOfSuperresolution;
typedef struct _InitializationParameterOfDetector InitializationParameterOfSegmentation;
typedef struct _InitializationParameterOfDetector InitializationParameterOfNLP;
typedef struct _InitializationParameterOfDetector InitializationParameterOfOcr;
typedef struct _InitializationParameterOfDetector InitializationParameterOfDB;
typedef struct _InitializationParameterOfDetector InitializationParameterOfSVTR;
}
#endif
Src/Utility/CommonUtility.cpp deleted 100644 → 0
View file @ 5a36110c
#include <CommonUtility.h>
#include <assert.h>
#include <ctype.h>
#include <time.h>
#include <stdlib.h>
#include <algorithm>
#include <sstream>
#include <vector>
#ifdef _WIN32
#include <io.h>
#include <direct.h>
#include <Windows.h>
#else
#include <unistd.h>
#include <dirent.h>
#include <sys/stat.h>
#include <sys/time.h>
#endif
#include <SimpleLog.h>
namespace migraphxSamples
{
_Time GetCurrentTime3()
{
_Time currentTime;
#if (defined WIN32 || defined _WIN32)
SYSTEMTIME systemTime;
GetLocalTime(&systemTime);
char temp[8] = { 0 };
sprintf(temp, "%04d", systemTime.wYear);
currentTime.year=string(temp);
sprintf(temp, "%02d", systemTime.wMonth);
currentTime.month=string(temp);
sprintf(temp, "%02d", systemTime.wDay);
currentTime.day=string(temp);
sprintf(temp, "%02d", systemTime.wHour);
currentTime.hour=string(temp);
sprintf(temp, "%02d", systemTime.wMinute);
currentTime.minute=string(temp);
sprintf(temp, "%02d", systemTime.wSecond);
currentTime.second=string(temp);
sprintf(temp, "%03d", systemTime.wMilliseconds);
currentTime.millisecond=string(temp);
sprintf(temp, "%d", systemTime.wDayOfWeek);
currentTime.weekDay=string(temp);
#else
struct timeval tv;
struct tm *p;
gettimeofday(&tv, NULL);
p = localtime(&tv.tv_sec);
char temp[8]={0};
sprintf(temp,"%04d",1900+p->tm_year);
currentTime.year=string(temp);
sprintf(temp,"%02d",1+p->tm_mon);
currentTime.month=string(temp);
sprintf(temp,"%02d",p->tm_mday);
currentTime.day=string(temp);
sprintf(temp,"%02d",p->tm_hour);
currentTime.hour=string(temp);
sprintf(temp,"%02d",p->tm_min);
currentTime.minute=string(temp);
sprintf(temp,"%02d",p->tm_sec);
currentTime.second=string(temp);
sprintf(temp,"%03d",tv.tv_usec/1000);
currentTime.millisecond = string(temp);
sprintf(temp, "%03d", tv.tv_usec % 1000);
currentTime.microsecond = string(temp);
sprintf(temp, "%d", p->tm_wday);
currentTime.weekDay = string(temp);
#endif
return currentTime;
}
std::vector<std::string> SplitString(std::string str, std::string separator)
{
std::string::size_type pos;
std::vector<std::string> result;
str+=separator;//扩展字符串以方便操作
int size=str.size();
for(int i=0; i<size; i++)
{
pos=str.find(separator,i);
if(pos<size)
{
std::string s=str.substr(i,pos-i);
result.push_back(s);
i=pos+separator.size()-1;
}
}
return result;
}
bool CompareConfidence(const ResultOfDetection &L,const ResultOfDetection &R)
{
return L.confidence > R.confidence;
}
bool CompareArea(const ResultOfDetection &L,const ResultOfDetection &R)
{
return L.boundingBox.area() > R.boundingBox.area();
}
void NMS(vector<ResultOfDetection> &detections, float IOUThreshold)
{
// sort
std::sort(detections.begin(), detections.end(), CompareConfidence);
for (int i = 0; i<detections.size(); ++i)
{
if (detections[i].exist)
{
for (int j = i + 1; j<detections.size(); ++j)
{
if (detections[j].exist)
{
// compute IOU
float intersectionArea = (detections[i].boundingBox & detections[j].boundingBox).area();
float intersectionRate = intersectionArea / (detections[i].boundingBox.area() + detections[j].boundingBox.area() - intersectionArea);
if (intersectionRate>IOUThreshold)
{
detections[j].exist = false;
}
}
}
}
}
}
}
Src/Utility/CommonUtility.h deleted 100644 → 0
View file @ 5a36110c
// 常用工具
#ifndef __COMMON_UTILITY_H__
#define __COMMON_UTILITY_H__
#include <mutex>
#include <string>
#include <vector>
#include <CommonDefinition.h>
using namespace std;
namespace migraphxSamples
{
// 分割字符串
std::vector<std::string> SplitString(std::string str,std::string separator);
// 排序规则: 按照置信度或者按照面积排序
bool CompareConfidence(const ResultOfDetection &L,const ResultOfDetection &R);
bool CompareArea(const ResultOfDetection &L,const ResultOfDetection &R);
void NMS(std::vector<ResultOfDetection> &detections, float IOUThreshold);
}
#endif
Src/Utility/Filesystem.cpp
View file @ b2735663
......@@ -11,12 +11,7 @@
#include <unistd.h>
#include <dirent.h>
#endif
#include <CommonUtility.h>
#include <opencv2/opencv.hpp>
#include <SimpleLog.h>
using namespace cv;
// 路径分隔符(Linux:‘/’,Windows:’\\’)
#ifdef _WIN32
#define PATH_SEPARATOR '\\'
......@@ -24,9 +19,31 @@ using namespace cv;
#define PATH_SEPARATOR '/'
#endif
using namespace std;
namespace migraphxSamples
{
static std::vector<std::string> SplitString(std::string str, std::string separator)
{
std::string::size_type pos;
std::vector<std::string> result;
str+=separator;//扩展字符串以方便操作
int size=str.size();
for(int i=0; i<size; i++)
{
pos=str.find(separator,i);
if(pos<size)
{
std::string s=str.substr(i,pos-i);
result.push_back(s);
i=pos+separator.size()-1;
}
}
return result;
}
#if defined _WIN32 || defined WINCE
const char dir_separators[] = "/\\";
......@@ -293,7 +310,7 @@ namespace migraphxSamples
}
else
{
LOG_INFO(stdout, "could not open directory: %s", directory.c_str());
printf("could not open directory: %s", directory.c_str());
}
}
......@@ -390,7 +407,7 @@ namespace migraphxSamples
#endif
if (!result)
{
LOG_INFO(stdout, "can't remove directory: %s\n", path.c_str());
printf("can't remove directory: %s\n", path.c_str());
}
}
else
......@@ -402,7 +419,7 @@ namespace migraphxSamples
#endif
if (!result)
{
LOG_INFO(stdout, "can't remove file: %s\n", path.c_str());
printf("can't remove file: %s\n", path.c_str());
}
}
}
......@@ -438,7 +455,7 @@ namespace migraphxSamples
{
RemoveAll(path);
++numberOfFiles;
LOG_INFO(stdout, "%s deleted! number of deleted files:%d\n", path.c_str(), numberOfFiles);
printf("%s deleted! number of deleted files:%d\n", path.c_str(), numberOfFiles);
}
}
......@@ -452,7 +469,7 @@ namespace migraphxSamples
}
else
{
LOG_INFO(stdout, "could not open directory: %s", directory.c_str());
printf("could not open directory: %s", directory.c_str());
}
// ����RemoveAllɾ��Ŀ¼
......@@ -592,17 +609,17 @@ namespace migraphxSamples
if(!srcFile.is_open())
{
LOG_ERROR(stdout,"can not open %s\n",srcPath.c_str());
printf("can not open %s\n",srcPath.c_str());
return false;
}
if(!dstFile.is_open())
{
LOG_ERROR(stdout, "can not open %s\n", dstPath.c_str());
printf("can not open %s\n", dstPath.c_str());
return false;
}
if(srcPath==dstPath)
{
LOG_ERROR(stdout, "src can not be same with dst\n");
printf("src can not be same with dst\n");
return false;
}
char buffer[2048];
......@@ -622,7 +639,7 @@ namespace migraphxSamples
{
if(srcPath==dstPath)
{
LOG_ERROR(stdout, "src can not be same with dst\n");
printf("src can not be same with dst\n");
return false;
}
......@@ -662,9 +679,9 @@ namespace migraphxSamples
// process
double process = (1.0*(i + 1) / fileNameList.size()) * 100;
LOG_INFO(stdout, "%s done! %f% \n", GetFileName(fileNameList[i]).c_str(), process);
printf("%s done! %f% \n", GetFileName(fileNameList[i]).c_str(), process);
}
LOG_INFO(stdout, "all done!(the number of files:%d)\n", fileNameList.size());
printf("all done!(the number of files:%d)\n", fileNameList.size());
return true;
......
Src/Utility/Filesystem.h
View file @ b2735663
......@@ -3,10 +3,8 @@
#ifndef __FILE_SYSTEM_H__
#define __FILE_SYSTEM_H__
#include <vector>
#include <string>
using namespace std;
#include <vector>
namespace migraphxSamples
{
......@@ -21,7 +19,7 @@ bool IsDirectory(const std::string &path);
bool IsPathSeparator(char c);
// 路径拼接
string JoinPath(const std::string &base, const std::string &path);
std::string JoinPath(const std::string &base, const std::string &path);
// 创建多级目录,注意：创建多级目录的时候，目标目录是不能有文件存在的
bool CreateDirectories(const std::string &directoryPath);
......@@ -49,14 +47,13 @@ void Remove(const std::string &directory, const std::string &extension="");
/** 获取路径的文件名和扩展名
*
* 示例：path为D:/1/1.txt,则GetFileName()为1.txt,GetFileName_NoExtension()为1,GetExtension()为.txt,GetParentPath()为D:/1/
*/
string GetFileName(const std::string &path); // 1.txt
string GetFileName_NoExtension(const std::string &path); // 1
string GetExtension(const std::string &path);// .txt
string GetParentPath(const std::string &path);// D:/1/
std::string GetFileName(const std::string &path);
std::string GetFileName_NoExtension(const std::string &path);
std::string GetExtension(const std::string &path);
std::string GetParentPath(const std::string &path);
// 拷贝文件:CopyFile("D:/1.txt","D:/2.txt");将1.txt拷贝为2.txt
// 拷贝文件
bool CopyFile(const std::string srcPath,const std::string dstPath);
/** 拷贝目录
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment