Update requirements.txt

[fix]Update requirements.txt For fix

.gitattributes

@@ -0,0 +1,5 @@
+*.h linguist-detectable=false
+*.cpp linguist-detectable=false
+*.tex linguist-detectable=false
+*.cs linguist-detectable=false
+*.tps linguist-detectable=false

.github/ISSUE_TEMPLATE/bug_report.md

@@ -0,0 +1,19 @@
+---
+name: Bug report
+about: Create a report to help us improve
+title: ''
+labels: ''
+assignees: ''
+
+---
+
+**Describe the bug 简述**
+
+**Screen Shot 截图**
+
+**Terminal Traceback 终端traceback（如果有）**
+
+
+Before submitting an issue 提交issue之前：
+- Please try to upgrade your code. 如果您的代码不是最新的，建议您先尝试更新代码
+- Please check project wiki for common problem solutions.项目[wiki](https://github.com/binary-husky/chatgpt_academic/wiki)有一些常见问题的解决方法

.github/ISSUE_TEMPLATE/feature_request.md

@@ -0,0 +1,10 @@
+---
+name: Feature request
+about: Suggest an idea for this project
+title: ''
+labels: ''
+assignees: ''
+
+---
+
+

.gitignore

@@ -131,6 +131,14 @@ dmypy.json
 # Pyre type checker
 .pyre/
 
+.vscode
+
 history
 ssr_conf
 config_private.py
+gpt_log
+private.md
+private_upload
+other_llms
+cradle*
+debug*

Dockerfile

@@ -1,17 +1,14 @@
-FROM ubuntu:latest
-
-RUN apt-get update && \
-    apt-get install -y python3 python3-pip && \
-    rm -rf /var/lib/apt/lists/*
+FROM python:3.11
 
 RUN echo '[global]' > /etc/pip.conf && \
     echo 'index-url = https://mirrors.aliyun.com/pypi/simple/' >> /etc/pip.conf && \
     echo 'trusted-host = mirrors.aliyun.com' >> /etc/pip.conf
 
-RUN pip3 install gradio requests[socks] mdtex2html
 
-COPY . /gpt
 WORKDIR /gpt
+COPY requirements.txt .
+RUN pip3 install -r requirements.txt
 
+COPY . .
 
-CMD ["python3", "main.py"]
+CMD ["python3", "-u", "main.py"]

Dockerfile+ChatGLM

@@ -0,0 +1,50 @@
+# How to build | 如何构建: docker build -t gpt-academic --network=host  -f Dockerfile+ChatGLM .
+# How to run | 如何运行 (1) 直接运行: docker run --rm -it --net=host --gpus=all gpt-academic
+# How to run | 如何运行 (2) 我想运行之前进容器做一些调整: docker run --rm -it --net=host --gpus=all gpt-academic bash
+ 
+# 从NVIDIA源，从而支持显卡运损（检查宿主的nvidia-smi中的cuda版本必须>=11.3）
+FROM nvidia/cuda:11.3.1-runtime-ubuntu20.04
+ARG useProxyNetwork=''
+RUN apt-get update
+RUN apt-get install -y curl proxychains curl 
+RUN apt-get install -y git python python3 python-dev python3-dev --fix-missing
+
+# 配置代理网络（构建Docker镜像时使用）
+# # comment out below if you do not need proxy network | 如果不需要翻墙 - 从此行向下删除
+RUN $useProxyNetwork curl cip.cc
+RUN sed -i '$ d' /etc/proxychains.conf
+RUN sed -i '$ d' /etc/proxychains.conf
+RUN echo "socks5 127.0.0.1 10880" >> /etc/proxychains.conf
+ARG useProxyNetwork=proxychains
+# # comment out above if you do not need proxy network | 如果不需要翻墙 - 从此行向上删除
+
+
+# use python3 as the system default python
+RUN curl -sS https://bootstrap.pypa.io/get-pip.py | python3.8
+
+# 下载分支
+WORKDIR /gpt
+RUN $useProxyNetwork git clone https://github.com/binary-husky/chatgpt_academic.git -b v3.0
+WORKDIR /gpt/chatgpt_academic
+RUN $useProxyNetwork python3 -m pip install -r requirements.txt
+RUN $useProxyNetwork python3 -m pip install -r request_llm/requirements_chatglm.txt
+RUN $useProxyNetwork python3 -m pip install torch --extra-index-url https://download.pytorch.org/whl/cu113
+
+# 预热CHATGLM参数（非必要 可选步骤）
+RUN echo ' \n\
+from transformers import AutoModel, AutoTokenizer \n\
+chatglm_tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True) \n\
+chatglm_model = AutoModel.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True).float() ' >> warm_up_chatglm.py
+RUN python3 -u warm_up_chatglm.py
+RUN $useProxyNetwork git pull
+
+# 为chatgpt-academic配置代理和API-KEY （非必要 可选步骤）
+RUN echo ' \n\
+API_KEY = "sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx" \n\
+USE_PROXY = True \n\
+LLM_MODEL = "chatglm" \n\
+LOCAL_MODEL_DEVICE = "cuda" \n\
+proxies = { "http": "socks5h://localhost:10880", "https": "socks5h://localhost:10880", } ' >> config_private.py
+
+# 启动
+CMD ["python3", "-u", "main.py"]

README.md

@@ -1,26 +1,298 @@
 # ChatGPT 学术优化
 
-**如果喜欢这个项目，请给它一个Star**
+**如果喜欢这个项目，请给它一个Star；如果你发明了更好用的快捷键或函数插件，欢迎发issue或者pull requests（dev分支）**
 
-## 使用docker
+If you like this project, please give it a Star. If you've come up with more useful academic shortcuts or functional plugins, feel free to open an issue or pull request （to `dev` branch）.
+
+> **Note**
+>
+> 1.请注意只有**红颜色**标识的函数插件（按钮）才支持读取文件，部分插件位于插件区的**下拉菜单**中。另外我们以**最高优先级**欢迎和处理任何新插件的PR！
+>
+> 2.本项目中每个文件的功能都在自译解[`self_analysis.md`](https://github.com/binary-husky/chatgpt_academic/wiki/chatgpt-academic%E9%A1%B9%E7%9B%AE%E8%87%AA%E8%AF%91%E8%A7%A3%E6%8A%A5%E5%91%8A)详细说明。随着版本的迭代，您也可以随时自行点击相关函数插件，调用GPT重新生成项目的自我解析报告。常见问题汇总在[`wiki`](https://github.com/binary-husky/chatgpt_academic/wiki/%E5%B8%B8%E8%A7%81%E9%97%AE%E9%A2%98)当中。
+> 
+> 3.如果您不太习惯部分中文命名的函数、注释或者界面，您可以随时点击相关函数插件，调用ChatGPT一键生成纯英文的项目源代码。
+>
+
+
+<div align="center">
+    
+功能 | 描述
+--- | ---
+一键润色 | 支持一键润色、一键查找论文语法错误
+一键中英互译 | 一键中英互译
+一键代码解释 | 可以正确显示代码、解释代码
+[自定义快捷键](https://www.bilibili.com/video/BV14s4y1E7jN) | 支持自定义快捷键
+[配置代理服务器](https://www.bilibili.com/video/BV1rc411W7Dr) | 支持配置代理服务器
+模块化设计 | 支持自定义高阶的实验性功能与[函数插件]，插件支持[热更新](https://github.com/binary-husky/chatgpt_academic/wiki/%E5%87%BD%E6%95%B0%E6%8F%92%E4%BB%B6%E6%8C%87%E5%8D%97)
+[自我程序剖析](https://www.bilibili.com/video/BV1cj411A7VW) | [函数插件] [一键读懂](https://github.com/binary-husky/chatgpt_academic/wiki/chatgpt-academic%E9%A1%B9%E7%9B%AE%E8%87%AA%E8%AF%91%E8%A7%A3%E6%8A%A5%E5%91%8A)本项目的源代码
+[程序剖析](https://www.bilibili.com/video/BV1cj411A7VW) | [函数插件] 一键可以剖析其他Python/C/C++/Java/Golang/Lua/Rect项目树
+读论文 | [函数插件] 一键解读latex论文全文并生成摘要
+Latex全文翻译、润色 | [函数插件] 一键翻译或润色latex论文
+批量注释生成 | [函数插件] 一键批量生成函数注释
+chat分析报告生成 | [函数插件] 运行后自动生成总结汇报
+[arxiv小助手](https://www.bilibili.com/video/BV1LM4y1279X) | [函数插件] 输入arxiv文章url即可一键翻译摘要+下载PDF
+[PDF论文全文翻译功能](https://www.bilibili.com/video/BV1KT411x7Wn) | [函数插件] PDF论文提取题目&摘要+翻译全文（多线程）
+[谷歌学术统合小助手](https://www.bilibili.com/video/BV19L411U7ia) (Version>=2.45) | [函数插件] 给定任意谷歌学术搜索页面URL，让gpt帮你选择有趣的文章
+公式显示 | 可以同时显示公式的tex形式和渲染形式
+图片显示 | 可以在markdown中显示图片
+多线程函数插件支持 | 支持多线调用chatgpt，一键处理海量文本或程序
+支持GPT输出的markdown表格 | 可以输出支持GPT的markdown表格
+启动暗色gradio[主题](https://github.com/binary-husky/chatgpt_academic/issues/173) | 在浏览器url后面添加```/?__dark-theme=true```可以切换dark主题
+huggingface免科学上网[在线体验](https://huggingface.co/spaces/qingxu98/gpt-academic) | 登陆huggingface后复制[此空间](https://huggingface.co/spaces/qingxu98/gpt-academic)
+[多LLM模型](https://www.bilibili.com/video/BV1EM411K7VH/)混合支持（[v3.0分支](https://github.com/binary-husky/chatgpt_academic/tree/v3.0)测试中） | 同时被ChatGPT和[清华ChatGLM](https://github.com/THUDM/ChatGLM-6B)伺候的感觉一定会很不错吧？
+兼容[TGUI](https://github.com/oobabooga/text-generation-webui)接入更多样的语言模型 | 接入opt-1.3b, galactica-1.3b等模型（[v3.0分支](https://github.com/binary-husky/chatgpt_academic/tree/v3.0)测试中）
+…… | ……
+
+</div>
+
+<!-- - 新界面（左：master主分支, 右：dev开发前沿） -->
+- 新界面（修改config.py中的LAYOUT选项即可实现“左右布局”和“上下布局”的切换）
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/230361456-61078362-a966-4eb5-b49e-3c62ef18b860.gif" width="700" >
+</div>
+
+
+
+- 所有按钮都通过读取functional.py动态生成，可随意加自定义功能，解放粘贴板
+<div align="center">
+<img src="img/公式.gif" width="700" >
+</div>
+
+- 润色/纠错
+<div align="center">
+<img src="img/润色.gif" width="700" >
+</div>
+
+
+- 支持GPT输出的markdown表格
+<div align="center">
+<img src="img/demo2.jpg" width="500" >
+</div>
+
+- 如果输出包含公式，会同时以tex形式和渲染形式显示，方便复制和阅读
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/230598842-1d7fcddd-815d-40ee-af60-baf488a199df.png" width="700" >
+</div>
+
+
+
+- 懒得看项目代码？整个工程直接给chatgpt炫嘴里
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226935232-6b6a73ce-8900-4aee-93f9-733c7e6fef53.png" width="700" >
+</div>
+
+- 多种大语言模型混合调用（[v3.0分支](https://github.com/binary-husky/chatgpt_academic/tree/v3.0)测试中）
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/231222778-34776885-a7f0-4f2c-b5f4-7cc2ef3ecb58.png" width="700" >
+</div>
+
+
+
+## 直接运行 (Windows, Linux or MacOS)
+
+### 1. 下载项目
+```sh
+git clone https://github.com/binary-husky/chatgpt_academic.git
+cd chatgpt_academic
+```
+
+### 2. 配置API_KEY和代理设置
+
+在`config.py`中，配置 海外Proxy 和 OpenAI API KEY，说明如下
+```
+1. 如果你在国内，需要设置海外代理才能够顺利使用 OpenAI API，设置方法请仔细阅读config.py（1.修改其中的USE_PROXY为True; 2.按照说明修改其中的proxies）。
+2. 配置 OpenAI API KEY。你需要在 OpenAI 官网上注册并获取 API KEY。一旦你拿到了 API KEY，在 config.py 文件里配置好即可。
+3. 与代理网络有关的issue（网络超时、代理不起作用）汇总到 https://github.com/binary-husky/chatgpt_academic/issues/1
+```
+（P.S. 程序运行时会优先检查是否存在名为`config_private.py`的私密配置文件，并用其中的配置覆盖`config.py`的同名配置。因此，如果您能理解我们的配置读取逻辑，我们强烈建议您在`config.py`旁边创建一个名为`config_private.py`的新配置文件，并把`config.py`中的配置转移（复制）到`config_private.py`中。`config_private.py`不受git管控，可以让您的隐私信息更加安全。）
+
+
+### 3. 安装依赖
+```sh
+# （选择一）推荐
+python -m pip install -r requirements.txt   
+
+# （选择二）如果您使用anaconda，步骤也是类似的：
+# （选择二.1）conda create -n gptac_venv python=3.11
+# （选择二.2）conda activate gptac_venv
+# （选择二.3）python -m pip install -r requirements.txt
+
+# 备注：使用官方pip源或者阿里pip源，其他pip源（如一些大学的pip）有可能出问题，临时换源方法： 
+# python -m pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/
+```
+
+### 4. 运行
+```sh
+python main.py
+```
+
+### 5. 测试实验性功能
+```
+- 测试C++项目头文件分析
+    input区域 输入 `./crazy_functions/test_project/cpp/libJPG` ， 然后点击 "[实验] 解析整个C++项目（input输入项目根路径）"
+- 测试给Latex项目写摘要
+    input区域 输入 `./crazy_functions/test_project/latex/attention` ， 然后点击 "[实验] 读tex论文写摘要（input输入项目根路径）"
+- 测试Python项目分析
+    input区域 输入 `./crazy_functions/test_project/python/dqn` ， 然后点击 "[实验] 解析整个py项目（input输入项目根路径）"
+- 测试自我代码解读
+    点击 "[实验] 请解析并解构此项目本身"
+- 测试实验功能模板函数（要求gpt回答历史上的今天发生了什么），您可以根据此函数为模板，实现更复杂的功能
+    点击 "[实验] 实验功能函数模板"
+```
+
+## 使用docker (Linux)
 
 ``` sh
 # 下载项目
 git clone https://github.com/binary-husky/chatgpt_academic.git
 cd chatgpt_academic
 # 配置 海外Proxy 和 OpenAI API KEY
-config.py
+用任意文本编辑器编辑 config.py
 # 安装
 docker build -t gpt-academic .
 # 运行
 docker run --rm -it --net=host gpt-academic
 
+# 测试实验性功能
+## 测试自我代码解读
+点击 "[实验] 请解析并解构此项目本身"
+## 测试实验功能模板函数（要求gpt回答历史上的今天发生了什么），您可以根据此函数为模板，实现更复杂的功能
+点击 "[实验] 实验功能函数模板"
+##（请注意在docker中运行时，需要额外注意程序的文件访问权限问题）
+## 测试C++项目头文件分析
+input区域 输入 ./crazy_functions/test_project/cpp/libJPG ， 然后点击 "[实验] 解析整个C++项目（input输入项目根路径）"
+## 测试给Latex项目写摘要
+input区域 输入 ./crazy_functions/test_project/latex/attention ， 然后点击 "[实验] 读tex论文写摘要（input输入项目根路径）"
+## 测试Python项目分析
+input区域 输入 ./crazy_functions/test_project/python/dqn ， 然后点击 "[实验] 解析整个py项目（input输入项目根路径）"
+
 ```
 
-## 参考项目
+## 其他部署方式
+- 使用WSL2（Windows Subsystem for Linux 子系统）
+请访问[部署wiki-1](https://github.com/binary-husky/chatgpt_academic/wiki/%E4%BD%BF%E7%94%A8WSL2%EF%BC%88Windows-Subsystem-for-Linux-%E5%AD%90%E7%B3%BB%E7%BB%9F%EF%BC%89%E9%83%A8%E7%BD%B2)
+
+- nginx远程部署
+请访问[部署wiki-2](https://github.com/binary-husky/chatgpt_academic/wiki/%E8%BF%9C%E7%A8%8B%E9%83%A8%E7%BD%B2%E7%9A%84%E6%8C%87%E5%AF%BC)
+
+
+## 自定义新的便捷按钮（学术快捷键自定义）
+打开functional.py，添加条目如下，然后重启程序即可。（如果按钮已经添加成功并可见，那么前缀、后缀都支持热修改，无需重启程序即可生效。）
+例如
 ```
-https://github.com/Python-Markdown/markdown
-https://github.com/gradio-app/gradio
-https://github.com/polarwinkel/mdtex2html
+"超级英译中": {
+
+    # 前缀，会被加在你的输入之前。例如，用来描述你的要求，例如翻译、解释代码、润色等等
+    "Prefix": "请翻译把下面一段内容成中文，然后用一个markdown表格逐一解释文中出现的专有名词：\n\n", 
+    
+    # 后缀，会被加在你的输入之后。例如，配合前缀可以把你的输入内容用引号圈起来。
+    "Suffix": "",
+    
+},
+```
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226899272-477c2134-ed71-4326-810c-29891fe4a508.png" width="500" >
+</div>
+
+
+如果你发明了更好用的学术快捷键，欢迎发issue或者pull requests！
+
+## 配置代理
+### 方法一：常规方法
+在```config.py```中修改端口与代理软件对应
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226571294-37a47cd9-4d40-4c16-97a2-d360845406f7.png" width="500" >
+<img src="https://user-images.githubusercontent.com/96192199/226838985-e5c95956-69c2-4c23-a4dd-cd7944eeb451.png" width="500" >
+</div>
+
+配置完成后，你可以用以下命令测试代理是否工作，如果一切正常，下面的代码将输出你的代理服务器所在地：
+```
+python check_proxy.py
+```
+### 方法二：纯新手教程
+[纯新手教程](https://github.com/binary-husky/chatgpt_academic/wiki/%E4%BB%A3%E7%90%86%E8%BD%AF%E4%BB%B6%E9%97%AE%E9%A2%98%E7%9A%84%E6%96%B0%E6%89%8B%E8%A7%A3%E5%86%B3%E6%96%B9%E6%B3%95%EF%BC%88%E6%96%B9%E6%B3%95%E5%8F%AA%E9%80%82%E7%94%A8%E4%BA%8E%E6%96%B0%E6%89%8B%EF%BC%89)
+
+## 兼容性测试
+
+### 图片显示：
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/228737599-bf0a9d9c-1808-4f43-ae15-dfcc7af0f295.png" width="800" >
+</div>
+
+
+### 如果一个程序能够读懂并剖析自己：
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226936850-c77d7183-0749-4c1c-9875-fd4891842d0c.png" width="800" >
+</div>
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226936618-9b487e4b-ab5b-4b6e-84c6-16942102e917.png" width="800" >
+</div>
+
+### 其他任意Python/Cpp项目剖析：
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226935232-6b6a73ce-8900-4aee-93f9-733c7e6fef53.png" width="800" >
+</div>
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226969067-968a27c1-1b9c-486b-8b81-ab2de8d3f88a.png" width="800" >
+</div>
+
+### Latex论文一键阅读理解与摘要生成
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/227504406-86ab97cd-f208-41c3-8e4a-7000e51cf980.png" width="800" >
+</div>
+
+### 自动报告生成
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/227503770-fe29ce2c-53fd-47b0-b0ff-93805f0c2ff4.png" height="300" >
+<img src="https://user-images.githubusercontent.com/96192199/227504617-7a497bb3-0a2a-4b50-9a8a-95ae60ea7afd.png" height="300" >
+<img src="https://user-images.githubusercontent.com/96192199/227504005-efeaefe0-b687-49d0-bf95-2d7b7e66c348.png" height="300" >
+</div>
+
+### 模块化功能设计
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/229288270-093643c1-0018-487a-81e6-1d7809b6e90f.png" height="400" >
+<img src="https://user-images.githubusercontent.com/96192199/227504931-19955f78-45cd-4d1c-adac-e71e50957915.png" height="400" >
+</div>
+
+
+### 源代码转译英文
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/229720562-fe6c3508-6142-4635-a83d-21eb3669baee.png" height="400" >
+</div>
+
+## Todo 与 版本规划:
+
+- version 3 (Todo): 
+- - 支持gpt4和其他更多llm
+- version 2.4+ (Todo): 
+- - 总结大工程源代码时文本过长、token溢出的问题
+- - 实现项目打包部署
+- - 函数插件参数接口优化
+- - 自更新
+- version 2.4: (1)新增PDF全文翻译功能; (2)新增输入区切换位置的功能; (3)新增垂直布局选项; (4)多线程函数插件优化。
+- version 2.3: 增强多线程交互性
+- version 2.2: 函数插件支持热重载
+- version 2.1: 可折叠式布局
+- version 2.0: 引入模块化函数插件
+- version 1.0: 基础功能
+
+## 参考与学习
+
+
+```
+代码中参考了很多其他优秀项目中的设计，主要包括：
+
+# 借鉴项目1：借鉴了ChuanhuChatGPT中读取OpenAI json的方法、记录历史问询记录的方法以及gradio queue的使用技巧
 https://github.com/GaiZhenbiao/ChuanhuChatGPT
+
+# 借鉴项目2：
+https://github.com/THUDM/ChatGLM-6B
+
 ```

check_proxy.py

@@ -0,0 +1,142 @@
+
+def check_proxy(proxies):
+    import requests
+    proxies_https = proxies['https'] if proxies is not None else '无'
+    try:
+        response = requests.get("https://ipapi.co/json/",
+                                proxies=proxies, timeout=4)
+        data = response.json()
+        print(f'查询代理的地理位置，返回的结果是{data}')
+        if 'country_name' in data:
+            country = data['country_name']
+            result = f"代理配置 {proxies_https}, 代理所在地：{country}"
+        elif 'error' in data:
+            result = f"代理配置 {proxies_https}, 代理所在地：未知，IP查询频率受限"
+        print(result)
+        return result
+    except:
+        result = f"代理配置 {proxies_https}, 代理所在地查询超时，代理可能无效"
+        print(result)
+        return result
+
+
+def backup_and_download(current_version, remote_version):
+    """
+    一键更新协议：备份和下载
+    """
+    from toolbox import get_conf
+    import shutil
+    import os
+    import requests
+    import zipfile
+    os.makedirs(f'./history', exist_ok=True)
+    backup_dir = f'./history/backup-{current_version}/'
+    new_version_dir = f'./history/new-version-{remote_version}/'
+    if os.path.exists(new_version_dir):
+        return new_version_dir
+    os.makedirs(new_version_dir)
+    shutil.copytree('./', backup_dir, ignore=lambda x, y: ['history'])
+    proxies, = get_conf('proxies')
+    r = requests.get(
+        'https://github.com/binary-husky/chatgpt_academic/archive/refs/heads/master.zip', proxies=proxies, stream=True)
+    zip_file_path = backup_dir+'/master.zip'
+    with open(zip_file_path, 'wb+') as f:
+        f.write(r.content)
+    dst_path = new_version_dir
+    with zipfile.ZipFile(zip_file_path, "r") as zip_ref:
+        for zip_info in zip_ref.infolist():
+            dst_file_path = os.path.join(dst_path, zip_info.filename)
+            if os.path.exists(dst_file_path):
+                os.remove(dst_file_path)
+            zip_ref.extract(zip_info, dst_path)
+    return new_version_dir
+
+
+def patch_and_restart(path):
+    """
+    一键更新协议：覆盖和重启
+    """
+    import distutils
+    import shutil
+    import os
+    import sys
+    import time
+    from colorful import print亮黄, print亮绿, print亮红
+    # if not using config_private, move origin config.py as config_private.py
+    if not os.path.exists('config_private.py'):
+        print亮黄('由于您没有设置config_private.py私密配置，现将您的现有配置移动至config_private.py以防止配置丢失，',
+              '另外您可以随时在history子文件夹下找回旧版的程序。')
+        shutil.copyfile('config.py', 'config_private.py')
+    distutils.dir_util.copy_tree(path+'/chatgpt_academic-master', './')
+    import subprocess
+    print亮绿('代码已经更新，即将更新pip包依赖……')
+    for i in reversed(range(5)): time.sleep(1); print(i)
+    try: 
+        subprocess.check_call([sys.executable, '-m', 'pip', 'install', '-r', 'requirements.txt'])
+    except:
+        print亮红('pip包依赖安装出现问题，需要手动安装新增的依赖库 `python -m pip install -r requirements.txt`，然后在用常规的`python main.py`的方式启动。')
+    print亮绿('更新完成，您可以随时在history子文件夹下找回旧版的程序，5s之后重启')
+    print亮红('假如重启失败，您可能需要手动安装新增的依赖库 `python -m pip install -r requirements.txt`，然后在用常规的`python main.py`的方式启动。')
+    print(' ------------------------------ -----------------------------------')
+    for i in reversed(range(8)): time.sleep(1); print(i)
+    os.execl(sys.executable, sys.executable, *sys.argv)
+
+
+def get_current_version():
+    import json
+    try:
+        with open('./version', 'r', encoding='utf8') as f:
+            current_version = json.loads(f.read())['version']
+    except:
+        current_version = ""
+    return current_version
+
+
+def auto_update():
+    """
+    一键更新协议：查询版本和用户意见
+    """
+    try:
+        from toolbox import get_conf
+        import requests
+        import time
+        import json
+        proxies, = get_conf('proxies')
+        response = requests.get(
+            "https://raw.githubusercontent.com/binary-husky/chatgpt_academic/master/version", proxies=proxies, timeout=1)
+        remote_json_data = json.loads(response.text)
+        remote_version = remote_json_data['version']
+        if remote_json_data["show_feature"]:
+            new_feature = "新功能：" + remote_json_data["new_feature"]
+        else:
+            new_feature = ""
+        with open('./version', 'r', encoding='utf8') as f:
+            current_version = f.read()
+            current_version = json.loads(current_version)['version']
+        if (remote_version - current_version) >= 0.01:
+            from colorful import print亮黄
+            print亮黄(
+                f'\n新版本可用。新版本:{remote_version}，当前版本:{current_version}。{new_feature}')
+            print('（1）Github更新地址:\nhttps://github.com/binary-husky/chatgpt_academic\n')
+            user_instruction = input('（2）是否一键更新代码（Y+回车=确认，输入其他/无输入+回车=不更新）？')
+            if user_instruction in ['Y', 'y']:
+                path = backup_and_download(current_version, remote_version)
+                try:
+                    patch_and_restart(path)
+                except:
+                    print('更新失败。')
+            else:
+                print('自动更新程序：已禁用')
+                return
+        else:
+            return
+    except:
+        print('自动更新程序：已禁用')
+
+
+if __name__ == '__main__':
+    import os
+    os.environ['no_proxy'] = '*'  # 避免代理网络产生意外污染
+    from toolbox import get_conf
+    proxies, = get_conf('proxies')
+    check_proxy(proxies)

colorful.py

@@ -0,0 +1,91 @@
+import platform
+from sys import stdout
+
+if platform.system()=="Linux":
+    pass
+else: 
+    from colorama import init
+    init()
+
+# Do you like the elegance of Chinese characters?
+def print红(*kw,**kargs):
+    print("\033[0;31m",*kw,"\033[0m",**kargs)
+def print绿(*kw,**kargs):
+    print("\033[0;32m",*kw,"\033[0m",**kargs)
+def print黄(*kw,**kargs):
+    print("\033[0;33m",*kw,"\033[0m",**kargs)
+def print蓝(*kw,**kargs):
+    print("\033[0;34m",*kw,"\033[0m",**kargs)
+def print紫(*kw,**kargs):
+    print("\033[0;35m",*kw,"\033[0m",**kargs)
+def print靛(*kw,**kargs):
+    print("\033[0;36m",*kw,"\033[0m",**kargs)
+
+def print亮红(*kw,**kargs):
+    print("\033[1;31m",*kw,"\033[0m",**kargs)
+def print亮绿(*kw,**kargs):
+    print("\033[1;32m",*kw,"\033[0m",**kargs)
+def print亮黄(*kw,**kargs):
+    print("\033[1;33m",*kw,"\033[0m",**kargs)
+def print亮蓝(*kw,**kargs):
+    print("\033[1;34m",*kw,"\033[0m",**kargs)
+def print亮紫(*kw,**kargs):
+    print("\033[1;35m",*kw,"\033[0m",**kargs)
+def print亮靛(*kw,**kargs):
+    print("\033[1;36m",*kw,"\033[0m",**kargs)
+
+
+
+def print亮红(*kw,**kargs):
+    print("\033[1;31m",*kw,"\033[0m",**kargs)
+def print亮绿(*kw,**kargs):
+    print("\033[1;32m",*kw,"\033[0m",**kargs)
+def print亮黄(*kw,**kargs):
+    print("\033[1;33m",*kw,"\033[0m",**kargs)
+def print亮蓝(*kw,**kargs):
+    print("\033[1;34m",*kw,"\033[0m",**kargs)
+def print亮紫(*kw,**kargs):
+    print("\033[1;35m",*kw,"\033[0m",**kargs)
+def print亮靛(*kw,**kargs):
+    print("\033[1;36m",*kw,"\033[0m",**kargs)
+    
+print_red = print红
+print_green = print绿
+print_yellow = print黄
+print_blue = print蓝
+print_purple = print紫
+print_indigo = print靛
+
+print_bold_red = print亮红
+print_bold_green = print亮绿
+print_bold_yellow = print亮黄
+print_bold_blue = print亮蓝
+print_bold_purple = print亮紫
+print_bold_indigo = print亮靛
+    
+if not stdout.isatty():
+    # redirection, avoid a fucked up log file
+    print红 = print
+    print绿 = print
+    print黄 = print
+    print蓝 = print
+    print紫 = print
+    print靛 = print
+    print亮红 = print
+    print亮绿 = print
+    print亮黄 = print
+    print亮蓝 = print
+    print亮紫 = print
+    print亮靛 = print
+    print_red = print
+    print_green = print
+    print_yellow = print
+    print_blue = print
+    print_purple = print
+    print_indigo = print
+    print_bold_red = print
+    print_bold_green = print
+    print_bold_yellow = print
+    print_bold_blue = print
+    print_bold_purple = print
+    print_bold_indigo = print

config.py

@@ -1,11 +1,53 @@
-# my_api_key = "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"
-API_KEY = "sk-此处填API秘钥"
-API_URL = "https://api.openai.com/v1/chat/completions"
+# [step 1]>> 例如： API_KEY = "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" （此key无效）
+API_KEY = "sk-此处填API密钥"
 
+# [step 2]>> 改为True应用代理，如果直接在海外服务器部署，此处不修改
 USE_PROXY = False
 if USE_PROXY:
-    proxies = { "http": "socks5h://localhost:11284", "https": "socks5h://localhost:11284", } 
-    print('网络代理状态：运行。')
+    # 填写格式是 [协议]://  [地址] :[端口]，填写之前不要忘记把USE_PROXY改成True，如果直接在海外服务器部署，此处不修改
+    # 例如    "socks5h://localhost:11284"
+    # [协议] 常见协议无非socks5h/http; 例如 v2**y 和 ss* 的默认本地协议是socks5h; 而cl**h 的默认本地协议是http
+    # [地址] 懂的都懂，不懂就填localhost或者127.0.0.1肯定错不了（localhost意思是代理软件安装在本机上）
+    # [端口] 在代理软件的设置里找。虽然不同的代理软件界面不一样，但端口号都应该在最显眼的位置上
+
+    # 代理网络的地址，打开你的科学上网软件查看代理的协议(socks5/http)、地址(localhost)和端口(11284)
+    proxies = {
+        #          [协议]://  [地址]  :[端口]
+        "http":  "socks5h://localhost:11284",
+        "https": "socks5h://localhost:11284",
+    }
 else:
     proxies = None
-    print('网络代理状态：未配置。无代理状态下很可能无法访问。')
+
+
+# [step 3]>> 以下配置可以优化体验，但大部分场合下并不需要修改
+# 对话窗的高度
+CHATBOT_HEIGHT = 1115
+
+# 代码高亮
+CODE_HIGHLIGHT = True
+
+# 窗口布局
+LAYOUT = "LEFT-RIGHT"  # "LEFT-RIGHT"（左右布局） # "TOP-DOWN"（上下布局）
+
+# 发送请求到OpenAI后，等待多久判定为超时
+TIMEOUT_SECONDS = 30
+
+# 网页的端口, -1代表随机端口
+WEB_PORT = -1
+
+# 如果OpenAI不响应（网络卡顿、代理失败、KEY失效），重试的次数限制
+MAX_RETRY = 2
+
+# OpenAI模型选择是（gpt4现在只对申请成功的人开放）
+LLM_MODEL = "gpt-3.5-turbo"
+
+# OpenAI的API_URL
+API_URL = "https://api.openai.com/v1/chat/completions"
+
+# 设置并行使用的线程数
+CONCURRENT_COUNT = 100
+
+# 设置用户名和密码（相关功能不稳定，与gradio版本和网络都相关，如果本地使用不建议加这个）
+# [("username", "password"), ("username2", "password2"), ...]
+AUTHENTICATION = []

core_functional.py

@@ -0,0 +1,71 @@
+# 'primary' 颜色对应 theme.py 中的 primary_hue
+# 'secondary' 颜色对应 theme.py 中的 neutral_hue
+# 'stop' 颜色对应 theme.py 中的 color_er
+# 默认按钮颜色是 secondary
+from toolbox import clear_line_break
+
+
+def get_core_functions():
+    return {
+        "英语学术润色": {
+            # 前言
+            "Prefix":   r"Below is a paragraph from an academic paper. Polish the writing to meet the academic style, " +
+                        r"improve the spelling, grammar, clarity, concision and overall readability. When necessary, rewrite the whole sentence. " +
+                        r"Furthermore, list all modification and explain the reasons to do so in markdown table." + "\n\n",
+            # 后语
+            "Suffix":   r"",
+            "Color":    r"secondary",    # 按钮颜色
+        },
+        "中文学术润色": {
+            "Prefix":   r"作为一名中文学术论文写作改进助理，你的任务是改进所提供文本的拼写、语法、清晰、简洁和整体可读性，" +
+                        r"同时分解长句，减少重复，并提供改进建议。请只提供文本的更正版本，避免包括解释。请编辑以下文本" + "\n\n",
+            "Suffix":   r"",
+        },
+        "查找语法错误": {
+            "Prefix":   r"Can you help me ensure that the grammar and the spelling is correct? " +
+                        r"Do not try to polish the text, if no mistake is found, tell me that this paragraph is good." +
+                        r"If you find grammar or spelling mistakes, please list mistakes you find in a two-column markdown table, " +
+                        r"put the original text the first column, " +
+                        r"put the corrected text in the second column and highlight the key words you fixed.""\n"
+                        r"Example:""\n"
+                        r"Paragraph: How is you? Do you knows what is it?""\n"
+                        r"| Original sentence | Corrected sentence |""\n"
+                        r"| :--- | :--- |""\n"
+                        r"| How **is** you? | How **are** you? |""\n"
+                        r"| Do you **knows** what **is** **it**? | Do you **know** what **it** **is** ? |""\n"
+                        r"Below is a paragraph from an academic paper. "
+                        r"You need to report all grammar and spelling mistakes as the example before."
+                        + "\n\n",
+            "Suffix":   r"",
+            "PreProcess": clear_line_break,    # 预处理：清除换行符
+        },
+        "中译英": {
+            "Prefix":   r"Please translate following sentence to English:" + "\n\n",
+            "Suffix":   r"",
+        },
+        "学术中英互译": {
+            "Prefix":   r"I want you to act as a scientific English-Chinese translator, " +
+                        r"I will provide you with some paragraphs in one language " +
+                        r"and your task is to accurately and academically translate the paragraphs only into the other language. " +
+                        r"Do not repeat the original provided paragraphs after translation. " +
+                        r"You should use artificial intelligence tools, " +
+                        r"such as natural language processing, and rhetorical knowledge " +
+                        r"and experience about effective writing techniques to reply. " +
+                        r"I'll give you my paragraphs as follows, tell me what language it is written in, and then translate:" + "\n\n",
+            "Suffix": "",
+            "Color": "secondary",
+        },
+        "英译中": {
+            "Prefix":   r"翻译成地道的中文：" + "\n\n",
+            "Suffix":   r"",
+        },
+        "找图片": {
+            "Prefix":   r"我需要你找一张网络图片。使用Unsplash API(https://source.unsplash.com/960x640/?<英语关键词>)获取图片URL，" +
+                        r"然后请使用Markdown格式封装，并且不要有反斜线，不要用代码块。现在，请按以下描述给我发送图片：" + "\n\n",
+            "Suffix":   r"",
+        },
+        "解释代码": {
+            "Prefix":   r"请解释以下代码：" + "\n```\n",
+            "Suffix":   "\n```\n",
+        },
+    }

crazy_functional.py

@@ -0,0 +1,169 @@
+from toolbox import HotReload  # HotReload 的意思是热更新，修改函数插件后，不需要重启程序，代码直接生效
+
+
+def get_crazy_functions():
+    ###################### 第一组插件 ###########################
+    # [第一组插件]: 最早期编写的项目插件和一些demo
+    from crazy_functions.读文章写摘要 import 读文章写摘要
+    from crazy_functions.生成函数注释 import 批量生成函数注释
+    from crazy_functions.解析项目源代码 import 解析项目本身
+    from crazy_functions.解析项目源代码 import 解析一个Python项目
+    from crazy_functions.解析项目源代码 import 解析一个C项目的头文件
+    from crazy_functions.解析项目源代码 import 解析一个C项目
+    from crazy_functions.解析项目源代码 import 解析一个Golang项目
+    from crazy_functions.解析项目源代码 import 解析一个Java项目
+    from crazy_functions.解析项目源代码 import 解析一个Rect项目
+    from crazy_functions.高级功能函数模板 import 高阶功能模板函数
+    from crazy_functions.代码重写为全英文_多线程 import 全项目切换英文
+    from crazy_functions.Latex全文润色 import Latex英文润色
+
+    function_plugins = {
+
+        "解析整个Python项目": {
+            "Color": "stop",    # 按钮颜色
+            "Function": HotReload(解析一个Python项目)
+        },
+        "解析整个C++项目头文件": {
+            "Color": "stop",    # 按钮颜色
+            "Function": HotReload(解析一个C项目的头文件)
+        },
+        "解析整个C++项目（.cpp/.hpp/.c/.h）": {
+            "Color": "stop",    # 按钮颜色
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(解析一个C项目)
+        },
+        "解析整个Go项目": {
+            "Color": "stop",    # 按钮颜色
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(解析一个Golang项目)
+        },
+        "解析整个Java项目": {
+            "Color": "stop",  # 按钮颜色
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(解析一个Java项目)
+        },
+        "解析整个React项目": {
+            "Color": "stop",  # 按钮颜色
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(解析一个Rect项目)
+        },
+        "读Tex论文写摘要": {
+            "Color": "stop",    # 按钮颜色
+            "Function": HotReload(读文章写摘要)
+        },
+        "批量生成函数注释": {
+            "Color": "stop",    # 按钮颜色
+            "Function": HotReload(批量生成函数注释)
+        },
+        "[多线程Demo] 解析此项目本身（源码自译解）": {
+            "Function": HotReload(解析项目本身)
+        },
+        "[多线程demo] 把本项目源代码切换成全英文": {
+            # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(全项目切换英文)
+        },
+        "[函数插件模板Demo] 历史上的今天": {
+            # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
+            "Function": HotReload(高阶功能模板函数)
+        },
+
+    }
+    ###################### 第二组插件 ###########################
+    # [第二组插件]: 经过充分测试，但功能上距离达到完美状态还差一点点
+    from crazy_functions.批量总结PDF文档 import 批量总结PDF文档
+    from crazy_functions.批量总结PDF文档pdfminer import 批量总结PDF文档pdfminer
+    from crazy_functions.总结word文档 import 总结word文档
+    from crazy_functions.批量翻译PDF文档_多线程 import 批量翻译PDF文档
+    from crazy_functions.谷歌检索小助手 import 谷歌检索小助手
+    from crazy_functions.理解PDF文档内容 import 理解PDF文档内容标准文件输入
+    from crazy_functions.Latex全文润色 import Latex中文润色
+    from crazy_functions.Latex全文翻译 import Latex中译英
+    from crazy_functions.Latex全文翻译 import Latex英译中
+
+    function_plugins.update({
+        "批量翻译PDF文档（多线程）": {
+            "Color": "stop",
+            "AsButton": True,  # 加入下拉菜单中
+            "Function": HotReload(批量翻译PDF文档)
+        },
+        "[测试功能] 批量总结PDF文档": {
+            "Color": "stop",
+            "AsButton": False,  # 加入下拉菜单中
+            # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
+            "Function": HotReload(批量总结PDF文档)
+        },
+        "[测试功能] 批量总结PDF文档pdfminer": {
+            "Color": "stop",
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(批量总结PDF文档pdfminer)
+        },
+        "谷歌学术检索助手（输入谷歌学术搜索页url）": {
+            "Color": "stop",
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(谷歌检索小助手)
+        },
+        "批量总结Word文档": {
+            "Color": "stop",
+            "Function": HotReload(总结word文档)
+        },
+        "理解PDF文档内容 （模仿ChatPDF）": {
+            # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
+            "Color": "stop",
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(理解PDF文档内容标准文件输入)
+        },
+        "[测试功能] 英文Latex项目全文润色（输入路径或上传压缩包）": {
+            # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
+            "Color": "stop",
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(Latex英文润色)
+        },
+        "[测试功能] 中文Latex项目全文润色（输入路径或上传压缩包）": {
+            # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
+            "Color": "stop",
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(Latex中文润色)
+        },
+        "[测试功能] Latex项目全文中译英（输入路径或上传压缩包）": {
+            # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
+            "Color": "stop",
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(Latex中译英)
+        },
+        "[测试功能] Latex项目全文英译中（输入路径或上传压缩包）": {
+            # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
+            "Color": "stop",
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(Latex英译中)
+        },
+
+        
+    })
+
+    ###################### 第三组插件 ###########################
+    # [第三组插件]: 尚未充分测试的函数插件，放在这里
+    try:
+        from crazy_functions.下载arxiv论文翻译摘要 import 下载arxiv论文并翻译摘要
+        function_plugins.update({
+            "一键下载arxiv论文并翻译摘要（先在input输入编号，如1812.10695）": {
+                "Color": "stop",
+                "AsButton": False,  # 加入下拉菜单中
+                "Function": HotReload(下载arxiv论文并翻译摘要)
+            }
+        })
+
+    except Exception as err:
+        print(f'[下载arxiv论文并翻译摘要] 插件导入失败 {str(err)}')
+        
+    from crazy_functions.解析项目源代码 import 解析一个Lua项目
+    function_plugins.update({
+        "解析整个Lua项目": {
+            "Color": "stop",    # 按钮颜色
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": HotReload(解析一个Lua项目)
+        },
+    })        
+
+    ###################### 第n组插件 ###########################
+    return function_plugins

crazy_functions/Latex全文润色.py

@@ -0,0 +1,176 @@
+from toolbox import update_ui
+from toolbox import CatchException, report_execption, write_results_to_file
+fast_debug = False
+
+class PaperFileGroup():
+    def __init__(self):
+        self.file_paths = []
+        self.file_contents = []
+        self.sp_file_contents = []
+        self.sp_file_index = []
+        self.sp_file_tag = []
+
+        # count_token
+        import tiktoken
+        from toolbox import get_conf
+        enc = tiktoken.encoding_for_model(*get_conf('LLM_MODEL'))
+        def get_token_num(txt): return len(enc.encode(txt))
+        self.get_token_num = get_token_num
+
+    def run_file_split(self, max_token_limit=1900):
+        """
+        将长文本分离开来
+        """
+        for index, file_content in enumerate(self.file_contents):
+            if self.get_token_num(file_content) < max_token_limit:
+                self.sp_file_contents.append(file_content)
+                self.sp_file_index.append(index)
+                self.sp_file_tag.append(self.file_paths[index])
+            else:
+                from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
+                segments = breakdown_txt_to_satisfy_token_limit_for_pdf(file_content, self.get_token_num, max_token_limit)
+                for j, segment in enumerate(segments):
+                    self.sp_file_contents.append(segment)
+                    self.sp_file_index.append(index)
+                    self.sp_file_tag.append(self.file_paths[index] + f".part-{j}.tex")
+
+        print('Segmentation: done')
+
+def 多文件润色(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='en'):
+    import time, os, re
+    from .crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency
+
+
+    #  <-------- 读取Latex文件，删除其中的所有注释 ----------> 
+    pfg = PaperFileGroup()
+
+    for index, fp in enumerate(file_manifest):
+        with open(fp, 'r', encoding='utf-8', errors='replace') as f:
+            file_content = f.read()
+            # 定义注释的正则表达式
+            comment_pattern = r'%.*'
+            # 使用正则表达式查找注释，并替换为空字符串
+            clean_tex_content = re.sub(comment_pattern, '', file_content)
+            # 记录删除注释后的文本
+            pfg.file_paths.append(fp)
+            pfg.file_contents.append(clean_tex_content)
+
+    #  <-------- 拆分过长的latex文件 ----------> 
+    pfg.run_file_split(max_token_limit=1024)
+    n_split = len(pfg.sp_file_contents)
+
+    #  <-------- 抽取摘要 ----------> 
+    # if language == 'en':
+    #     abs_extract_inputs = f"Please write an abstract for this paper"
+
+    # # 单线，获取文章meta信息
+    # paper_meta_info = yield from request_gpt_model_in_new_thread_with_ui_alive(
+    #     inputs=abs_extract_inputs,
+    #     inputs_show_user=f"正在抽取摘要信息。",
+    #     llm_kwargs=llm_kwargs,
+    #     chatbot=chatbot, history=[],
+    #     sys_prompt="Your job is to collect information from materials。",
+    # )
+
+    #  <-------- 多线程润色开始 ----------> 
+    if language == 'en':
+        inputs_array = ["Below is a section from an academic paper, polish this section to meet the academic standard, improve the grammar, clarity and overall readability, do not modify any latex command such as \section, \cite and equations:" + 
+                        f"\n\n{frag}" for frag in pfg.sp_file_contents]
+        inputs_show_user_array = [f"Polish {f}" for f in pfg.sp_file_tag]
+        sys_prompt_array = ["You are a professional academic paper writer." for _ in range(n_split)]
+    elif language == 'zh':
+        inputs_array = [f"以下是一篇学术论文中的一段内容，请将此部分润色以满足学术标准，提高语法、清晰度和整体可读性，不要修改任何LaTeX命令，例如\section，\cite和方程式：" + 
+                        f"\n\n{frag}" for frag in pfg.sp_file_contents]
+        inputs_show_user_array = [f"润色 {f}" for f in pfg.sp_file_tag]
+        sys_prompt_array=["你是一位专业的中文学术论文作家。" for _ in range(n_split)]
+
+
+    gpt_response_collection = yield from request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
+        inputs_array=inputs_array,
+        inputs_show_user_array=inputs_show_user_array,
+        llm_kwargs=llm_kwargs,
+        chatbot=chatbot,
+        history_array=[[""] for _ in range(n_split)],
+        sys_prompt_array=sys_prompt_array,
+        max_workers=10,  # OpenAI所允许的最大并行过载
+        scroller_max_len = 80
+    )
+
+    #  <-------- 整理结果，退出 ----------> 
+    create_report_file_name = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) + f"-chatgpt.polish.md"
+    res = write_results_to_file(gpt_response_collection, file_name=create_report_file_name)
+    history = gpt_response_collection
+    chatbot.append((f"{fp}完成了吗？", res))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+
+@CatchException
+def Latex英文润色(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    # 基本信息：功能、贡献者
+    chatbot.append([
+        "函数插件功能？",
+        "对整个Latex项目进行润色。函数插件贡献者: Binary-Husky"])
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import tiktoken
+    except:
+        report_execption(chatbot, history,
+                         a=f"解析项目: {txt}",
+                         b=f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade tiktoken```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 多文件润色(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='en')
+
+
+
+
+
+
+@CatchException
+def Latex中文润色(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    # 基本信息：功能、贡献者
+    chatbot.append([
+        "函数插件功能？",
+        "对整个Latex项目进行润色。函数插件贡献者: Binary-Husky"])
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import tiktoken
+    except:
+        report_execption(chatbot, history,
+                         a=f"解析项目: {txt}",
+                         b=f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade tiktoken```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 多文件润色(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='zh')

crazy_functions/Latex全文翻译.py

@@ -0,0 +1,176 @@
+from toolbox import update_ui
+from toolbox import CatchException, report_execption, write_results_to_file
+fast_debug = False
+
+class PaperFileGroup():
+    def __init__(self):
+        self.file_paths = []
+        self.file_contents = []
+        self.sp_file_contents = []
+        self.sp_file_index = []
+        self.sp_file_tag = []
+
+        # count_token
+        import tiktoken
+        from toolbox import get_conf
+        enc = tiktoken.encoding_for_model(*get_conf('LLM_MODEL'))
+        def get_token_num(txt): return len(enc.encode(txt))
+        self.get_token_num = get_token_num
+
+    def run_file_split(self, max_token_limit=1900):
+        """
+        将长文本分离开来
+        """
+        for index, file_content in enumerate(self.file_contents):
+            if self.get_token_num(file_content) < max_token_limit:
+                self.sp_file_contents.append(file_content)
+                self.sp_file_index.append(index)
+                self.sp_file_tag.append(self.file_paths[index])
+            else:
+                from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
+                segments = breakdown_txt_to_satisfy_token_limit_for_pdf(file_content, self.get_token_num, max_token_limit)
+                for j, segment in enumerate(segments):
+                    self.sp_file_contents.append(segment)
+                    self.sp_file_index.append(index)
+                    self.sp_file_tag.append(self.file_paths[index] + f".part-{j}.tex")
+
+        print('Segmentation: done')
+
+def 多文件翻译(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='en'):
+    import time, os, re
+    from .crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency
+
+    #  <-------- 读取Latex文件，删除其中的所有注释 ----------> 
+    pfg = PaperFileGroup()
+
+    for index, fp in enumerate(file_manifest):
+        with open(fp, 'r', encoding='utf-8', errors='replace') as f:
+            file_content = f.read()
+            # 定义注释的正则表达式
+            comment_pattern = r'%.*'
+            # 使用正则表达式查找注释，并替换为空字符串
+            clean_tex_content = re.sub(comment_pattern, '', file_content)
+            # 记录删除注释后的文本
+            pfg.file_paths.append(fp)
+            pfg.file_contents.append(clean_tex_content)
+
+    #  <-------- 拆分过长的latex文件 ----------> 
+    pfg.run_file_split(max_token_limit=1024)
+    n_split = len(pfg.sp_file_contents)
+
+    #  <-------- 抽取摘要 ----------> 
+    # if language == 'en':
+    #     abs_extract_inputs = f"Please write an abstract for this paper"
+
+    # # 单线，获取文章meta信息
+    # paper_meta_info = yield from request_gpt_model_in_new_thread_with_ui_alive(
+    #     inputs=abs_extract_inputs,
+    #     inputs_show_user=f"正在抽取摘要信息。",
+    #     llm_kwargs=llm_kwargs,
+    #     chatbot=chatbot, history=[],
+    #     sys_prompt="Your job is to collect information from materials。",
+    # )
+
+    #  <-------- 多线程润色开始 ----------> 
+    if language == 'en->zh':
+        inputs_array = ["Below is a section from an English academic paper, translate it into Chinese, do not modify any latex command such as \section, \cite and equations:" + 
+                        f"\n\n{frag}" for frag in pfg.sp_file_contents]
+        inputs_show_user_array = [f"翻译 {f}" for f in pfg.sp_file_tag]
+        sys_prompt_array = ["You are a professional academic paper translator." for _ in range(n_split)]
+    elif language == 'zh->en':
+        inputs_array = [f"Below is a section from a Chinese academic paper, translate it into English, do not modify any latex command such as \section, \cite and equations:" + 
+                        f"\n\n{frag}" for frag in pfg.sp_file_contents]
+        inputs_show_user_array = [f"润色 {f}" for f in pfg.sp_file_tag]
+        sys_prompt_array = ["You are a professional academic paper translator." for _ in range(n_split)]
+
+    gpt_response_collection = yield from request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
+        inputs_array=inputs_array,
+        inputs_show_user_array=inputs_show_user_array,
+        llm_kwargs=llm_kwargs,
+        chatbot=chatbot,
+        history_array=[[""] for _ in range(n_split)],
+        sys_prompt_array=sys_prompt_array,
+        max_workers=10,  # OpenAI所允许的最大并行过载
+        scroller_max_len = 80
+    )
+
+    #  <-------- 整理结果，退出 ----------> 
+    create_report_file_name = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) + f"-chatgpt.polish.md"
+    res = write_results_to_file(gpt_response_collection, file_name=create_report_file_name)
+    history = gpt_response_collection
+    chatbot.append((f"{fp}完成了吗？", res))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+
+
+
+
+@CatchException
+def Latex英译中(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    # 基本信息：功能、贡献者
+    chatbot.append([
+        "函数插件功能？",
+        "对整个Latex项目进行翻译。函数插件贡献者: Binary-Husky"])
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import tiktoken
+    except:
+        report_execption(chatbot, history,
+                         a=f"解析项目: {txt}",
+                         b=f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade tiktoken```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 多文件翻译(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='en->zh')
+
+
+
+
+
+@CatchException
+def Latex中译英(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    # 基本信息：功能、贡献者
+    chatbot.append([
+        "函数插件功能？",
+        "对整个Latex项目进行翻译。函数插件贡献者: Binary-Husky"])
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import tiktoken
+    except:
+        report_execption(chatbot, history,
+                         a=f"解析项目: {txt}",
+                         b=f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade tiktoken```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 多文件翻译(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='zh->en')

crazy_functions/crazy_utils.py

@@ -0,0 +1,530 @@
+import traceback
+from toolbox import update_ui
+
+def input_clipping(inputs, history, max_token_limit):
+    import tiktoken
+    import numpy as np
+    from toolbox import get_conf
+    enc = tiktoken.encoding_for_model(*get_conf('LLM_MODEL'))
+    def get_token_num(txt): return len(enc.encode(txt))
+
+    mode = 'input-and-history'
+    # 当 输入部分的token占比 小于 全文的一半时，只裁剪历史
+    input_token_num = get_token_num(inputs)
+    if input_token_num < max_token_limit//2: 
+        mode = 'only-history'
+        max_token_limit = max_token_limit - input_token_num
+
+    everything = [inputs] if mode == 'input-and-history' else ['']
+    everything.extend(history)
+    n_token = get_token_num('\n'.join(everything))
+    everything_token = [get_token_num(e) for e in everything]
+    delta = max(everything_token) // 16 # 截断时的颗粒度
+        
+    while n_token > max_token_limit:
+        where = np.argmax(everything_token)
+        encoded = enc.encode(everything[where])
+        clipped_encoded = encoded[:len(encoded)-delta]
+        everything[where] = enc.decode(clipped_encoded)[:-1]    # -1 to remove the may-be illegal char
+        everything_token[where] = get_token_num(everything[where])
+        n_token = get_token_num('\n'.join(everything))
+
+    if mode == 'input-and-history':
+        inputs = everything[0]
+    else:
+        pass
+    history = everything[1:]
+    return inputs, history
+
+def request_gpt_model_in_new_thread_with_ui_alive(
+        inputs, inputs_show_user, llm_kwargs, 
+        chatbot, history, sys_prompt, refresh_interval=0.2,
+        handle_token_exceed=True, 
+        retry_times_at_unknown_error=2,
+        ):
+    """
+    Request GPT model，请求GPT模型同时维持用户界面活跃。
+
+    输入参数 Args （以_array结尾的输入变量都是列表，列表长度为子任务的数量，执行时，会把列表拆解，放到每个子线程中分别执行）:
+        inputs (string): List of inputs （输入）
+        inputs_show_user (string): List of inputs to show user（展现在报告中的输入，借助此参数，在汇总报告中隐藏啰嗦的真实输入，增强报告的可读性）
+        top_p (float): Top p value for sampling from model distribution （GPT参数，浮点数）
+        temperature (float): Temperature value for sampling from model distribution（GPT参数，浮点数）
+        chatbot: chatbot inputs and outputs （用户界面对话窗口句柄，用于数据流可视化）
+        history (list): List of chat history （历史，对话历史列表）
+        sys_prompt (string): List of system prompts （系统输入，列表，用于输入给GPT的前提提示，比如你是翻译官怎样怎样）
+        refresh_interval (float, optional): Refresh interval for UI (default: 0.2) （刷新时间间隔频率，建议低于1，不可高于3，仅仅服务于视觉效果）
+        handle_token_exceed：是否自动处理token溢出的情况，如果选择自动处理，则会在溢出时暴力截断，默认开启
+        retry_times_at_unknown_error：失败时的重试次数
+
+    输出 Returns:
+        future: 输出，GPT返回的结果
+    """
+    import time
+    from concurrent.futures import ThreadPoolExecutor
+    from request_llm.bridge_chatgpt import predict_no_ui_long_connection
+    # 用户反馈
+    chatbot.append([inputs_show_user, ""])
+    msg = '正常'
+    yield from update_ui(chatbot=chatbot, history=[]) # 刷新界面
+    executor = ThreadPoolExecutor(max_workers=16)
+    mutable = ["", time.time()]
+    def _req_gpt(inputs, history, sys_prompt):
+        retry_op = retry_times_at_unknown_error
+        exceeded_cnt = 0
+        while True:
+            try:
+                # 【第一种情况】：顺利完成
+                result = predict_no_ui_long_connection(
+                    inputs=inputs, llm_kwargs=llm_kwargs,
+                    history=history, sys_prompt=sys_prompt, observe_window=mutable)
+                return result
+            except ConnectionAbortedError as token_exceeded_error:
+                # 【第二种情况】：Token溢出
+                if handle_token_exceed:
+                    exceeded_cnt += 1
+                    # 【选择处理】 尝试计算比例，尽可能多地保留文本
+                    from toolbox import get_reduce_token_percent
+                    p_ratio, n_exceed = get_reduce_token_percent(str(token_exceeded_error))
+                    MAX_TOKEN = 4096
+                    EXCEED_ALLO = 512 + 512 * exceeded_cnt
+                    inputs, history = input_clipping(inputs, history, max_token_limit=MAX_TOKEN-EXCEED_ALLO)
+                    mutable[0] += f'[Local Message] 警告，文本过长将进行截断，Token溢出数：{n_exceed}。\n\n'
+                    continue # 返回重试
+                else:
+                    # 【选择放弃】
+                    tb_str = '```\n' + traceback.format_exc() + '```'
+                    mutable[0] += f"[Local Message] 警告，在执行过程中遭遇问题, Traceback：\n\n{tb_str}\n\n"
+                    return mutable[0] # 放弃
+            except:
+                # 【第三种情况】：其他错误：重试几次
+                tb_str = '```\n' + traceback.format_exc() + '```'
+                mutable[0] += f"[Local Message] 警告，在执行过程中遭遇问题, Traceback：\n\n{tb_str}\n\n"
+                if retry_op > 0: 
+                    retry_op -= 1
+                    mutable[0] += f"[Local Message] 重试中 {retry_times_at_unknown_error-retry_op}/{retry_times_at_unknown_error}：\n\n"
+                    time.sleep(5)
+                    continue # 返回重试
+                else:
+                    time.sleep(5)
+                    return mutable[0] # 放弃
+
+    future = executor.submit(_req_gpt, inputs, history, sys_prompt)
+    while True:
+        # yield一次以刷新前端页面
+        time.sleep(refresh_interval)
+        # “喂狗”（看门狗）
+        mutable[1] = time.time()
+        if future.done():
+            break
+        chatbot[-1] = [chatbot[-1][0], mutable[0]]
+        yield from update_ui(chatbot=chatbot, history=[]) # 刷新界面
+
+    final_result = future.result()
+    chatbot[-1] = [chatbot[-1][0], final_result]
+    yield from update_ui(chatbot=chatbot, history=[]) # 如果最后成功了，则删除报错信息
+    return final_result
+
+
+def request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
+        inputs_array, inputs_show_user_array, llm_kwargs, 
+        chatbot, history_array, sys_prompt_array, 
+        refresh_interval=0.2, max_workers=10, scroller_max_len=30,
+        handle_token_exceed=True, show_user_at_complete=False,
+        retry_times_at_unknown_error=2,
+        ):
+    """
+    Request GPT model using multiple threads with UI and high efficiency
+    请求GPT模型的[多线程]版。
+    具备以下功能：
+        实时在UI上反馈远程数据流
+        使用线程池，可调节线程池的大小避免openai的流量限制错误
+        处理中途中止的情况
+        网络等出问题时，会把traceback和已经接收的数据转入输出
+
+    输入参数 Args （以_array结尾的输入变量都是列表，列表长度为子任务的数量，执行时，会把列表拆解，放到每个子线程中分别执行）:
+        inputs_array (list): List of inputs （每个子任务的输入）
+        inputs_show_user_array (list): List of inputs to show user（每个子任务展现在报告中的输入，借助此参数，在汇总报告中隐藏啰嗦的真实输入，增强报告的可读性）
+        llm_kwargs: llm_kwargs参数
+        chatbot: chatbot （用户界面对话窗口句柄，用于数据流可视化）
+        history_array (list): List of chat history （历史对话输入，双层列表，第一层列表是子任务分解，第二层列表是对话历史）
+        sys_prompt_array (list): List of system prompts （系统输入，列表，用于输入给GPT的前提提示，比如你是翻译官怎样怎样）
+        refresh_interval (float, optional): Refresh interval for UI (default: 0.2) （刷新时间间隔频率，建议低于1，不可高于3，仅仅服务于视觉效果）
+        max_workers (int, optional): Maximum number of threads (default: 10) （最大线程数，如果子任务非常多，需要用此选项防止高频地请求openai导致错误）
+        scroller_max_len (int, optional): Maximum length for scroller (default: 30)（数据流的显示最后收到的多少个字符，仅仅服务于视觉效果）
+        handle_token_exceed (bool, optional): （是否在输入过长时，自动缩减文本）
+        handle_token_exceed：是否自动处理token溢出的情况，如果选择自动处理，则会在溢出时暴力截断，默认开启
+        show_user_at_complete (bool, optional): (在结束时，把完整输入-输出结果显示在聊天框)
+        retry_times_at_unknown_error：子任务失败时的重试次数
+
+    输出 Returns:
+        list: List of GPT model responses （每个子任务的输出汇总，如果某个子任务出错，response中会携带traceback报错信息，方便调试和定位问题。）
+    """
+    import time, random
+    from concurrent.futures import ThreadPoolExecutor
+    from request_llm.bridge_chatgpt import predict_no_ui_long_connection
+    assert len(inputs_array) == len(history_array)
+    assert len(inputs_array) == len(sys_prompt_array)
+    executor = ThreadPoolExecutor(max_workers=max_workers)
+    n_frag = len(inputs_array)
+    # 用户反馈
+    chatbot.append(["请开始多线程操作。", ""])
+    msg = '正常'
+    yield from update_ui(chatbot=chatbot, history=[]) # 刷新界面
+    # 异步原子
+    mutable = [["", time.time(), "等待中"] for _ in range(n_frag)]
+
+    def _req_gpt(index, inputs, history, sys_prompt):
+        gpt_say = ""
+        retry_op = retry_times_at_unknown_error
+        exceeded_cnt = 0
+        mutable[index][2] = "执行中"
+        while True:
+            try:
+                # 【第一种情况】：顺利完成
+                # time.sleep(10); raise RuntimeError("测试")
+                gpt_say = predict_no_ui_long_connection(
+                    inputs=inputs, llm_kwargs=llm_kwargs, history=history, 
+                    sys_prompt=sys_prompt, observe_window=mutable[index], console_slience=True
+                )
+                mutable[index][2] = "已成功"
+                return gpt_say
+            except ConnectionAbortedError as token_exceeded_error:
+                # 【第二种情况】：Token溢出，
+                if handle_token_exceed:
+                    exceeded_cnt += 1
+                    # 【选择处理】 尝试计算比例，尽可能多地保留文本
+                    from toolbox import get_reduce_token_percent
+                    p_ratio, n_exceed = get_reduce_token_percent(str(token_exceeded_error))
+                    MAX_TOKEN = 4096
+                    EXCEED_ALLO = 512 + 512 * exceeded_cnt
+                    inputs, history = input_clipping(inputs, history, max_token_limit=MAX_TOKEN-EXCEED_ALLO)
+                    gpt_say += f'[Local Message] 警告，文本过长将进行截断，Token溢出数：{n_exceed}。\n\n'
+                    mutable[index][2] = f"截断重试"
+                    continue # 返回重试
+                else:
+                    # 【选择放弃】
+                    tb_str = '```\n' + traceback.format_exc() + '```'
+                    gpt_say += f"[Local Message] 警告，线程{index}在执行过程中遭遇问题, Traceback：\n\n{tb_str}\n\n"
+                    if len(mutable[index][0]) > 0: gpt_say += "此线程失败前收到的回答：\n\n" + mutable[index][0]
+                    mutable[index][2] = "输入过长已放弃"
+                    return gpt_say # 放弃
+            except:
+                # 【第三种情况】：其他错误
+                tb_str = '```\n' + traceback.format_exc() + '```'
+                gpt_say += f"[Local Message] 警告，线程{index}在执行过程中遭遇问题, Traceback：\n\n{tb_str}\n\n"
+                if len(mutable[index][0]) > 0: gpt_say += "此线程失败前收到的回答：\n\n" + mutable[index][0]
+                if retry_op > 0: 
+                    retry_op -= 1
+                    wait = random.randint(5, 20)
+                    for i in range(wait):# 也许等待十几秒后，情况会好转
+                        mutable[index][2] = f"等待重试 {wait-i}"; time.sleep(1)
+                    mutable[index][2] = f"重试中 {retry_times_at_unknown_error-retry_op}/{retry_times_at_unknown_error}"
+                    continue # 返回重试
+                else:
+                    mutable[index][2] = "已失败"
+                    wait = 5
+                    time.sleep(5)
+                    return gpt_say # 放弃
+
+    # 异步任务开始
+    futures = [executor.submit(_req_gpt, index, inputs, history, sys_prompt) for index, inputs, history, sys_prompt in zip(
+        range(len(inputs_array)), inputs_array, history_array, sys_prompt_array)]
+    cnt = 0
+    while True:
+        # yield一次以刷新前端页面
+        time.sleep(refresh_interval)
+        cnt += 1
+        worker_done = [h.done() for h in futures]
+        if all(worker_done):
+            executor.shutdown()
+            break
+        # 更好的UI视觉效果
+        observe_win = []
+        # print([mutable[thread_index][2] for thread_index, _ in enumerate(worker_done)])
+        # 每个线程都要“喂狗”（看门狗）
+        for thread_index, _ in enumerate(worker_done):
+            mutable[thread_index][1] = time.time()
+        # 在前端打印些好玩的东西
+        for thread_index, _ in enumerate(worker_done):
+            print_something_really_funny = "[ ...`"+mutable[thread_index][0][-scroller_max_len:].\
+                replace('\n', '').replace('```', '...').replace(
+                    ' ', '.').replace('<br/>', '.....').replace('$', '.')+"`... ]"
+            observe_win.append(print_something_really_funny)
+        stat_str = ''.join([f'`{mutable[thread_index][2]}`: {obs}\n\n' 
+                            if not done else f'`{mutable[thread_index][2]}`\n\n' 
+                            for thread_index, done, obs in zip(range(len(worker_done)), worker_done, observe_win)])
+        chatbot[-1] = [chatbot[-1][0], f'多线程操作已经开始，完成情况: \n\n{stat_str}' + ''.join(['.']*(cnt % 10+1))]
+        msg = "正常"
+        yield from update_ui(chatbot=chatbot, history=[]) # 刷新界面
+    # 异步任务结束
+    gpt_response_collection = []
+    for inputs_show_user, f in zip(inputs_show_user_array, futures):
+        gpt_res = f.result()
+        gpt_response_collection.extend([inputs_show_user, gpt_res])
+
+    if show_user_at_complete:
+        for inputs_show_user, f in zip(inputs_show_user_array, futures):
+            gpt_res = f.result()
+            chatbot.append([inputs_show_user, gpt_res])
+            yield from update_ui(chatbot=chatbot, history=[]) # 刷新界面
+            time.sleep(1)
+    return gpt_response_collection
+
+
+def WithRetry(f):
+    """
+        装饰器函数，用于自动重试。
+    """
+    def decorated(retry, res_when_fail, *args, **kwargs):
+        assert retry >= 0
+        while True:
+            try:
+                res = yield from f(*args, **kwargs)
+                return res
+            except:
+                retry -= 1
+                if retry<0:
+                    print("达到最大重试次数")
+                    break
+                else:
+                    print("重试中……")
+                    continue
+        return res_when_fail
+    return decorated
+
+
+def breakdown_txt_to_satisfy_token_limit(txt, get_token_fn, limit):
+    def cut(txt_tocut, must_break_at_empty_line):  # 递归
+        if get_token_fn(txt_tocut) <= limit:
+            return [txt_tocut]
+        else:
+            lines = txt_tocut.split('\n')
+            estimated_line_cut = limit / get_token_fn(txt_tocut) * len(lines)
+            estimated_line_cut = int(estimated_line_cut)
+            for cnt in reversed(range(estimated_line_cut)):
+                if must_break_at_empty_line:
+                    if lines[cnt] != "":
+                        continue
+                print(cnt)
+                prev = "\n".join(lines[:cnt])
+                post = "\n".join(lines[cnt:])
+                if get_token_fn(prev) < limit:
+                    break
+            if cnt == 0:
+                print('what the fuck ?')
+                raise RuntimeError("存在一行极长的文本！")
+            # print(len(post))
+            # 列表递归接龙
+            result = [prev]
+            result.extend(cut(post, must_break_at_empty_line))
+            return result
+    try:
+        return cut(txt, must_break_at_empty_line=True)
+    except RuntimeError:
+        return cut(txt, must_break_at_empty_line=False)
+
+
+def breakdown_txt_to_satisfy_token_limit_for_pdf(txt, get_token_fn, limit):
+    def cut(txt_tocut, must_break_at_empty_line):  # 递归
+        if get_token_fn(txt_tocut) <= limit:
+            return [txt_tocut]
+        else:
+            lines = txt_tocut.split('\n')
+            estimated_line_cut = limit / get_token_fn(txt_tocut) * len(lines)
+            estimated_line_cut = int(estimated_line_cut)
+            cnt = 0
+            for cnt in reversed(range(estimated_line_cut)):
+                if must_break_at_empty_line:
+                    if lines[cnt] != "":
+                        continue
+                print(cnt)
+                prev = "\n".join(lines[:cnt])
+                post = "\n".join(lines[cnt:])
+                if get_token_fn(prev) < limit:
+                    break
+            if cnt == 0:
+                # print('what the fuck ? 存在一行极长的文本！')
+                raise RuntimeError("存在一行极长的文本！")
+            # print(len(post))
+            # 列表递归接龙
+            result = [prev]
+            result.extend(cut(post, must_break_at_empty_line))
+            return result
+    try:
+        return cut(txt, must_break_at_empty_line=True)
+    except RuntimeError:
+        try:
+            return cut(txt, must_break_at_empty_line=False)
+        except RuntimeError:
+            # 这个中文的句号是故意的，作为一个标识而存在
+            res = cut(txt.replace('.', '。\n'), must_break_at_empty_line=False)
+            return [r.replace('。\n', '.') for r in res]
+
+
+
+def read_and_clean_pdf_text(fp):
+    """
+    这个函数用于分割pdf，用了很多trick，逻辑较乱，效果奇好
+
+    **输入参数说明**
+    - `fp`：需要读取和清理文本的pdf文件路径
+
+    **输出参数说明**
+    - `meta_txt`：清理后的文本内容字符串
+    - `page_one_meta`：第一页清理后的文本内容列表
+
+    **函数功能**
+    读取pdf文件并清理其中的文本内容，清理规则包括：
+    - 提取所有块元的文本信息，并合并为一个字符串
+    - 去除短块（字符数小于100）并替换为回车符
+    - 清理多余的空行
+    - 合并小写字母开头的段落块并替换为空格
+    - 清除重复的换行
+    - 将每个换行符替换为两个换行符，使每个段落之间有两个换行符分隔
+    """
+    import fitz, copy
+    import re
+    import numpy as np
+    from colorful import print亮黄, print亮绿
+    fc = 0
+    fs = 1
+    fb = 2
+    REMOVE_FOOT_NOTE = True
+    REMOVE_FOOT_FFSIZE_PERCENT = 0.95 
+    def primary_ffsize(l):
+        fsize_statiscs = {}
+        for wtf in l['spans']:
+            if wtf['size'] not in fsize_statiscs: fsize_statiscs[wtf['size']] = 0
+            fsize_statiscs[wtf['size']] += len(wtf['text'])
+        return max(fsize_statiscs, key=fsize_statiscs.get)
+        
+    def ffsize_same(a,b):
+        return abs((a-b)/max(a,b)) < 0.02
+    # file_content = ""
+    with fitz.open(fp) as doc:
+        meta_txt = []
+        meta_font = []
+
+        meta_line = []
+        meta_span = []
+        for index, page in enumerate(doc):
+            # file_content += page.get_text()
+            text_areas = page.get_text("dict")  # 获取页面上的文本信息
+            for t in text_areas['blocks']:
+                if 'lines' in t:
+                    pf = 998
+                    for l in t['lines']:
+                        txt_line = "".join([wtf['text'] for wtf in l['spans']])
+                        pf = primary_ffsize(l)
+                        meta_line.append([txt_line, pf, l['bbox'], l])
+                        for wtf in l['spans']: # for l in t['lines']:
+                            meta_span.append([wtf['text'], wtf['size'], len(wtf['text'])])
+                    # meta_line.append(["NEW_BLOCK", pf])
+            # 块元提取                           for each word segment with in line                       for each line         cross-line words                          for each block
+            meta_txt.extend([" ".join(["".join([wtf['text'] for wtf in l['spans']]) for l in t['lines']]).replace(
+                '- ', '') for t in text_areas['blocks'] if 'lines' in t])
+            meta_font.extend([np.mean([np.mean([wtf['size'] for wtf in l['spans']])
+                             for l in t['lines']]) for t in text_areas['blocks'] if 'lines' in t])
+            if index == 0:
+                page_one_meta = [" ".join(["".join([wtf['text'] for wtf in l['spans']]) for l in t['lines']]).replace(
+                    '- ', '') for t in text_areas['blocks'] if 'lines' in t]
+        # 获取正文主字体
+        fsize_statiscs = {}
+        for span in meta_span:
+            if span[1] not in fsize_statiscs: fsize_statiscs[span[1]] = 0
+            fsize_statiscs[span[1]] += span[2]
+        main_fsize = max(fsize_statiscs, key=fsize_statiscs.get)
+        if REMOVE_FOOT_NOTE:
+            give_up_fize_threshold = main_fsize * REMOVE_FOOT_FFSIZE_PERCENT
+
+        # 切分和重新整合
+        mega_sec = []
+        sec = []
+        for index, line in enumerate(meta_line):
+            if index == 0: 
+                sec.append(line[fc])
+                continue
+            if REMOVE_FOOT_NOTE:
+                if meta_line[index][fs] <= give_up_fize_threshold:
+                    continue
+            if ffsize_same(meta_line[index][fs], meta_line[index-1][fs]):
+                # 尝试识别段落
+                if meta_line[index][fc].endswith('.') and\
+                    (meta_line[index-1][fc] != 'NEW_BLOCK') and \
+                    (meta_line[index][fb][2] - meta_line[index][fb][0]) < (meta_line[index-1][fb][2] - meta_line[index-1][fb][0]) * 0.7:
+                    sec[-1] += line[fc]
+                    sec[-1] += "\n\n"
+                else:
+                    sec[-1] += " "
+                    sec[-1] += line[fc]
+            else:
+                if (index+1 < len(meta_line)) and \
+                    meta_line[index][fs] > main_fsize:
+                    # 单行 + 字体大
+                    mega_sec.append(copy.deepcopy(sec))
+                    sec = []
+                    sec.append("# " + line[fc])
+                else:
+                    # 尝试识别section
+                    if meta_line[index-1][fs] > meta_line[index][fs]:
+                        sec.append("\n" + line[fc])
+                    else:
+                        sec.append(line[fc])
+        mega_sec.append(copy.deepcopy(sec))
+
+        finals = []
+        for ms in mega_sec:
+            final = " ".join(ms)
+            final = final.replace('- ', ' ')
+            finals.append(final)
+        meta_txt = finals
+
+        def 把字符太少的块清除为回车(meta_txt):
+            for index, block_txt in enumerate(meta_txt):
+                if len(block_txt) < 100:
+                    meta_txt[index] = '\n'
+            return meta_txt
+        meta_txt = 把字符太少的块清除为回车(meta_txt)
+
+        def 清理多余的空行(meta_txt):
+            for index in reversed(range(1, len(meta_txt))):
+                if meta_txt[index] == '\n' and meta_txt[index-1] == '\n':
+                    meta_txt.pop(index)
+            return meta_txt
+        meta_txt = 清理多余的空行(meta_txt)
+
+        def 合并小写开头的段落块(meta_txt):
+            def starts_with_lowercase_word(s):
+                pattern = r"^[a-z]+"
+                match = re.match(pattern, s)
+                if match:
+                    return True
+                else:
+                    return False
+            for _ in range(100):
+                for index, block_txt in enumerate(meta_txt):
+                    if starts_with_lowercase_word(block_txt):
+                        if meta_txt[index-1] != '\n':
+                            meta_txt[index-1] += ' '
+                        else:
+                            meta_txt[index-1] = ''
+                        meta_txt[index-1] += meta_txt[index]
+                        meta_txt[index] = '\n'
+            return meta_txt
+        meta_txt = 合并小写开头的段落块(meta_txt)
+        meta_txt = 清理多余的空行(meta_txt)
+
+        meta_txt = '\n'.join(meta_txt)
+        # 清除重复的换行
+        for _ in range(5):
+            meta_txt = meta_txt.replace('\n\n', '\n')
+
+        # 换行 -> 双换行
+        meta_txt = meta_txt.replace('\n', '\n\n')
+
+        for f in finals:
+            print亮黄(f)
+            print亮绿('***************************')
+
+    return meta_txt, page_one_meta

crazy_functions/test_project/cpp/cppipc/buffer.cpp

@@ -0,0 +1,87 @@
+#include "libipc/buffer.h"
+#include "libipc/utility/pimpl.h"
+
+#include <cstring>
+
+namespace ipc {
+
+bool operator==(buffer const & b1, buffer const & b2) {
+    return (b1.size() == b2.size()) && (std::memcmp(b1.data(), b2.data(), b1.size()) == 0);
+}
+
+bool operator!=(buffer const & b1, buffer const & b2) {
+    return !(b1 == b2);
+}
+
+class buffer::buffer_ : public pimpl<buffer_> {
+public:
+    void*       p_;
+    std::size_t s_;
+    void*       a_;
+    buffer::destructor_t d_;
+
+    buffer_(void* p, std::size_t s, buffer::destructor_t d, void* a)
+        : p_(p), s_(s), a_(a), d_(d) {
+    }
+
+    ~buffer_() {
+        if (d_ == nullptr) return;
+        d_((a_ == nullptr) ? p_ : a_, s_);
+    }
+};
+
+buffer::buffer()
+    : buffer(nullptr, 0, nullptr, nullptr) {
+}
+
+buffer::buffer(void* p, std::size_t s, destructor_t d)
+    : p_(p_->make(p, s, d, nullptr)) {
+}
+
+buffer::buffer(void* p, std::size_t s, destructor_t d, void* additional)
+    : p_(p_->make(p, s, d, additional)) {
+}
+
+buffer::buffer(void* p, std::size_t s)
+    : buffer(p, s, nullptr) {
+}
+
+buffer::buffer(char const & c)
+    : buffer(const_cast<char*>(&c), 1) {
+}
+
+buffer::buffer(buffer&& rhs)
+    : buffer() {
+    swap(rhs);
+}
+
+buffer::~buffer() {
+    p_->clear();
+}
+
+void buffer::swap(buffer& rhs) {
+    std::swap(p_, rhs.p_);
+}
+
+buffer& buffer::operator=(buffer rhs) {
+    swap(rhs);
+    return *this;
+}
+
+bool buffer::empty() const noexcept {
+    return (impl(p_)->p_ == nullptr) || (impl(p_)->s_ == 0);
+}
+
+void* buffer::data() noexcept {
+    return impl(p_)->p_;
+}
+
+void const * buffer::data() const noexcept {
+    return impl(p_)->p_;
+}
+
+std::size_t buffer::size() const noexcept {
+    return impl(p_)->s_;
+}
+
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/ipc.cpp

@@ -0,0 +1,701 @@
+
+#include <type_traits>
+#include <cstring>
+#include <algorithm>
+#include <utility>          // std::pair, std::move, std::forward
+#include <atomic>
+#include <type_traits>      // aligned_storage_t
+#include <string>
+#include <vector>
+#include <array>
+#include <cassert>
+
+#include "libipc/ipc.h"
+#include "libipc/def.h"
+#include "libipc/shm.h"
+#include "libipc/pool_alloc.h"
+#include "libipc/queue.h"
+#include "libipc/policy.h"
+#include "libipc/rw_lock.h"
+#include "libipc/waiter.h"
+
+#include "libipc/utility/log.h"
+#include "libipc/utility/id_pool.h"
+#include "libipc/utility/scope_guard.h"
+#include "libipc/utility/utility.h"
+
+#include "libipc/memory/resource.h"
+#include "libipc/platform/detail.h"
+#include "libipc/circ/elem_array.h"
+
+namespace {
+
+using msg_id_t = std::uint32_t;
+using acc_t    = std::atomic<msg_id_t>;
+
+template <std::size_t DataSize, std::size_t AlignSize>
+struct msg_t;
+
+template <std::size_t AlignSize>
+struct msg_t<0, AlignSize> {
+    msg_id_t     cc_id_;
+    msg_id_t     id_;
+    std::int32_t remain_;
+    bool         storage_;
+};
+
+template <std::size_t DataSize, std::size_t AlignSize>
+struct msg_t : msg_t<0, AlignSize> {
+    std::aligned_storage_t<DataSize, AlignSize> data_ {};
+
+    msg_t() = default;
+    msg_t(msg_id_t cc_id, msg_id_t id, std::int32_t remain, void const * data, std::size_t size)
+        : msg_t<0, AlignSize> {cc_id, id, remain, (data == nullptr) || (size == 0)} {
+        if (this->storage_) {
+            if (data != nullptr) {
+                // copy storage-id
+                *reinterpret_cast<ipc::storage_id_t*>(&data_) =
+                     *static_cast<ipc::storage_id_t const *>(data);
+            }
+        }
+        else std::memcpy(&data_, data, size);
+    }
+};
+
+template <typename T>
+ipc::buff_t make_cache(T& data, std::size_t size) {
+    auto ptr = ipc::mem::alloc(size);
+    std::memcpy(ptr, &data, (ipc::detail::min)(sizeof(data), size));
+    return { ptr, size, ipc::mem::free };
+}
+
+struct cache_t {
+    std::size_t fill_;
+    ipc::buff_t buff_;
+
+    cache_t(std::size_t f, ipc::buff_t && b)
+        : fill_(f), buff_(std::move(b))
+    {}
+
+    void append(void const * data, std::size_t size) {
+        if (fill_ >= buff_.size() || data == nullptr || size == 0) return;
+        auto new_fill = (ipc::detail::min)(fill_ + size, buff_.size());
+        std::memcpy(static_cast<ipc::byte_t*>(buff_.data()) + fill_, data, new_fill - fill_);
+        fill_ = new_fill;
+    }
+};
+
+auto cc_acc() {
+    static ipc::shm::handle acc_h("__CA_CONN__", sizeof(acc_t));
+    return static_cast<acc_t*>(acc_h.get());
+}
+
+IPC_CONSTEXPR_ std::size_t align_chunk_size(std::size_t size) noexcept {
+    return (((size - 1) / ipc::large_msg_align) + 1) * ipc::large_msg_align;
+}
+
+IPC_CONSTEXPR_ std::size_t calc_chunk_size(std::size_t size) noexcept {
+    return ipc::make_align(alignof(std::max_align_t), align_chunk_size(
+           ipc::make_align(alignof(std::max_align_t), sizeof(std::atomic<ipc::circ::cc_t>)) + size));
+}
+
+struct chunk_t {
+    std::atomic<ipc::circ::cc_t> &conns() noexcept {
+        return *reinterpret_cast<std::atomic<ipc::circ::cc_t> *>(this);
+    }
+
+    void *data() noexcept {
+        return reinterpret_cast<ipc::byte_t *>(this)
+             + ipc::make_align(alignof(std::max_align_t), sizeof(std::atomic<ipc::circ::cc_t>));
+    }
+};
+
+struct chunk_info_t {
+    ipc::id_pool<> pool_;
+    ipc::spin_lock lock_;
+
+    IPC_CONSTEXPR_ static std::size_t chunks_mem_size(std::size_t chunk_size) noexcept {
+        return ipc::id_pool<>::max_count * chunk_size;
+    }
+
+    ipc::byte_t *chunks_mem() noexcept {
+        return reinterpret_cast<ipc::byte_t *>(this + 1);
+    }
+
+    chunk_t *at(std::size_t chunk_size, ipc::storage_id_t id) noexcept {
+        if (id < 0) return nullptr;
+        return reinterpret_cast<chunk_t *>(chunks_mem() + (chunk_size * id));
+    }
+};
+
+auto& chunk_storages() {
+    class chunk_handle_t {
+        ipc::shm::handle handle_;
+
+    public:
+        chunk_info_t *get_info(std::size_t chunk_size) {
+            if (!handle_.valid() &&
+                !handle_.acquire( ("__CHUNK_INFO__" + ipc::to_string(chunk_size)).c_str(), 
+                                  sizeof(chunk_info_t) + chunk_info_t::chunks_mem_size(chunk_size) )) {
+                ipc::error("[chunk_storages] chunk_shm.id_info_.acquire failed: chunk_size = %zd\n", chunk_size);
+                return nullptr;
+            }
+            auto info = static_cast<chunk_info_t*>(handle_.get());
+            if (info == nullptr) {
+                ipc::error("[chunk_storages] chunk_shm.id_info_.get failed: chunk_size = %zd\n", chunk_size);
+                return nullptr;
+            }
+            return info;
+        }
+    };
+    static ipc::map<std::size_t, chunk_handle_t> chunk_hs;
+    return chunk_hs;
+}
+
+chunk_info_t *chunk_storage_info(std::size_t chunk_size) {
+    auto &storages = chunk_storages();
+    std::decay_t<decltype(storages)>::iterator it;
+    {
+        static ipc::rw_lock lock;
+        IPC_UNUSED_ std::shared_lock<ipc::rw_lock> guard {lock};
+        if ((it = storages.find(chunk_size)) == storages.end()) {
+            using chunk_handle_t = std::decay_t<decltype(storages)>::value_type::second_type;
+            guard.unlock();
+            IPC_UNUSED_ std::lock_guard<ipc::rw_lock> guard {lock};
+            it = storages.emplace(chunk_size, chunk_handle_t{}).first;
+        }
+    }
+    return it->second.get_info(chunk_size);
+}
+
+std::pair<ipc::storage_id_t, void*> acquire_storage(std::size_t size, ipc::circ::cc_t conns) {
+    std::size_t chunk_size = calc_chunk_size(size);
+    auto info = chunk_storage_info(chunk_size);
+    if (info == nullptr) return {};
+
+    info->lock_.lock();
+    info->pool_.prepare();
+    // got an unique id
+    auto id = info->pool_.acquire();
+    info->lock_.unlock();
+
+    auto chunk = info->at(chunk_size, id);
+    if (chunk == nullptr) return {};
+    chunk->conns().store(conns, std::memory_order_relaxed);
+    return { id, chunk->data() };
+}
+
+void *find_storage(ipc::storage_id_t id, std::size_t size) {
+    if (id < 0) {
+        ipc::error("[find_storage] id is invalid: id = %ld, size = %zd\n", (long)id, size);
+        return nullptr;
+    }
+    std::size_t chunk_size = calc_chunk_size(size);
+    auto info = chunk_storage_info(chunk_size);
+    if (info == nullptr) return nullptr;
+    return info->at(chunk_size, id)->data();
+}
+
+void release_storage(ipc::storage_id_t id, std::size_t size) {
+    if (id < 0) {
+        ipc::error("[release_storage] id is invalid: id = %ld, size = %zd\n", (long)id, size);
+        return;
+    }
+    std::size_t chunk_size = calc_chunk_size(size);
+    auto info = chunk_storage_info(chunk_size);
+    if (info == nullptr) return;
+    info->lock_.lock();
+    info->pool_.release(id);
+    info->lock_.unlock();
+}
+
+template <ipc::relat Rp, ipc::relat Rc>
+bool sub_rc(ipc::wr<Rp, Rc, ipc::trans::unicast>, 
+            std::atomic<ipc::circ::cc_t> &/*conns*/, ipc::circ::cc_t /*curr_conns*/, ipc::circ::cc_t /*conn_id*/) noexcept {
+    return true;
+}
+
+template <ipc::relat Rp, ipc::relat Rc>
+bool sub_rc(ipc::wr<Rp, Rc, ipc::trans::broadcast>, 
+            std::atomic<ipc::circ::cc_t> &conns, ipc::circ::cc_t curr_conns, ipc::circ::cc_t conn_id) noexcept {
+    auto last_conns = curr_conns & ~conn_id;
+    for (unsigned k = 0;;) {
+        auto chunk_conns  = conns.load(std::memory_order_acquire);
+        if (conns.compare_exchange_weak(chunk_conns, chunk_conns & last_conns, std::memory_order_release)) {
+            return (chunk_conns & last_conns) == 0;
+        }
+        ipc::yield(k);
+    }
+}
+
+template <typename Flag>
+void recycle_storage(ipc::storage_id_t id, std::size_t size, ipc::circ::cc_t curr_conns, ipc::circ::cc_t conn_id) {
+    if (id < 0) {
+        ipc::error("[recycle_storage] id is invalid: id = %ld, size = %zd\n", (long)id, size);
+        return;
+    }
+    std::size_t chunk_size = calc_chunk_size(size);
+    auto info = chunk_storage_info(chunk_size);
+    if (info == nullptr) return;
+
+    auto chunk = info->at(chunk_size, id);
+    if (chunk == nullptr) return;
+
+    if (!sub_rc(Flag{}, chunk->conns(), curr_conns, conn_id)) {
+        return;
+    }
+    info->lock_.lock();
+    info->pool_.release(id);
+    info->lock_.unlock();
+}
+
+template <typename MsgT>
+bool clear_message(void* p) {
+    auto msg = static_cast<MsgT*>(p);
+    if (msg->storage_) {
+        std::int32_t r_size = static_cast<std::int32_t>(ipc::data_length) + msg->remain_;
+        if (r_size <= 0) {
+            ipc::error("[clear_message] invalid msg size: %d\n", (int)r_size);
+            return true;
+        }
+        release_storage(
+            *reinterpret_cast<ipc::storage_id_t*>(&msg->data_),
+            static_cast<std::size_t>(r_size));
+    }
+    return true;
+}
+
+struct conn_info_head {
+
+    ipc::string name_;
+    msg_id_t    cc_id_; // connection-info id
+    ipc::detail::waiter cc_waiter_, wt_waiter_, rd_waiter_;
+    ipc::shm::handle acc_h_;
+
+    conn_info_head(char const * name)
+        : name_     {name}
+        , cc_id_    {(cc_acc() == nullptr) ? 0 : cc_acc()->fetch_add(1, std::memory_order_relaxed)}
+        , cc_waiter_{("__CC_CONN__" + name_).c_str()}
+        , wt_waiter_{("__WT_CONN__" + name_).c_str()}
+        , rd_waiter_{("__RD_CONN__" + name_).c_str()}
+        , acc_h_    {("__AC_CONN__" + name_).c_str(), sizeof(acc_t)} {
+    }
+
+    void quit_waiting() {
+        cc_waiter_.quit_waiting();
+        wt_waiter_.quit_waiting();
+        rd_waiter_.quit_waiting();
+    }
+
+    auto acc() {
+        return static_cast<acc_t*>(acc_h_.get());
+    }
+
+    auto& recv_cache() {
+        thread_local ipc::unordered_map<msg_id_t, cache_t> tls;
+        return tls;
+    }
+};
+
+template <typename W, typename F>
+bool wait_for(W& waiter, F&& pred, std::uint64_t tm) {
+    if (tm == 0) return !pred();
+    for (unsigned k = 0; pred();) {
+        bool ret = true;
+        ipc::sleep(k, [&k, &ret, &waiter, &pred, tm] {
+            ret = waiter.wait_if(std::forward<F>(pred), tm);
+            k   = 0;
+        });
+        if (!ret) return false; // timeout or fail
+        if (k == 0) break; // k has been reset
+    }
+    return true;
+}
+
+template <typename Policy,
+          std::size_t DataSize  = ipc::data_length,
+          std::size_t AlignSize = (ipc::detail::min)(DataSize, alignof(std::max_align_t))>
+struct queue_generator {
+
+    using queue_t = ipc::queue<msg_t<DataSize, AlignSize>, Policy>;
+
+    struct conn_info_t : conn_info_head {
+        queue_t que_;
+
+        conn_info_t(char const * name)
+            : conn_info_head{name}
+            , que_{("__QU_CONN__" +
+                    ipc::to_string(DataSize) + "__" +
+                    ipc::to_string(AlignSize) + "__" + name).c_str()} {
+        }
+
+        void disconnect_receiver() {
+            bool dis = que_.disconnect();
+            this->quit_waiting();
+            if (dis) {
+                this->recv_cache().clear();
+            }
+        }
+    };
+};
+
+template <typename Policy>
+struct detail_impl {
+
+using policy_t    = Policy;
+using flag_t      = typename policy_t::flag_t;
+using queue_t     = typename queue_generator<policy_t>::queue_t;
+using conn_info_t = typename queue_generator<policy_t>::conn_info_t;
+
+constexpr static conn_info_t* info_of(ipc::handle_t h) noexcept {
+    return static_cast<conn_info_t*>(h);
+}
+
+constexpr static queue_t* queue_of(ipc::handle_t h) noexcept {
+    return (info_of(h) == nullptr) ? nullptr : &(info_of(h)->que_);
+}
+
+/* API implementations */
+
+static void disconnect(ipc::handle_t h) {
+    auto que = queue_of(h);
+    if (que == nullptr) {
+        return;
+    }
+    que->shut_sending();
+    assert(info_of(h) != nullptr);
+    info_of(h)->disconnect_receiver();
+}
+
+static bool reconnect(ipc::handle_t * ph, bool start_to_recv) {
+    assert(ph != nullptr);
+    assert(*ph != nullptr);
+    auto que = queue_of(*ph);
+    if (que == nullptr) {
+        return false;
+    }
+    if (start_to_recv) {
+        que->shut_sending();
+        if (que->connect()) { // wouldn't connect twice
+            info_of(*ph)->cc_waiter_.broadcast();
+            return true;
+        }
+        return false;
+    }
+    // start_to_recv == false
+    if (que->connected()) {
+        info_of(*ph)->disconnect_receiver();
+    }
+    return que->ready_sending();
+}
+
+static bool connect(ipc::handle_t * ph, char const * name, bool start_to_recv) {
+    assert(ph != nullptr);
+    if (*ph == nullptr) {
+        *ph = ipc::mem::alloc<conn_info_t>(name);
+    }
+    return reconnect(ph, start_to_recv);
+}
+
+static void destroy(ipc::handle_t h) {
+    disconnect(h);
+    ipc::mem::free(info_of(h));
+}
+
+static std::size_t recv_count(ipc::handle_t h) noexcept {
+    auto que = queue_of(h);
+    if (que == nullptr) {
+        return ipc::invalid_value;
+    }
+    return que->conn_count();
+}
+
+static bool wait_for_recv(ipc::handle_t h, std::size_t r_count, std::uint64_t tm) {
+    auto que = queue_of(h);
+    if (que == nullptr) {
+        return false;
+    }
+    return wait_for(info_of(h)->cc_waiter_, [que, r_count] {
+        return que->conn_count() < r_count;
+    }, tm);
+}
+
+template <typename F>
+static bool send(F&& gen_push, ipc::handle_t h, void const * data, std::size_t size) {
+    if (data == nullptr || size == 0) {
+        ipc::error("fail: send(%p, %zd)\n", data, size);
+        return false;
+    }
+    auto que = queue_of(h);
+    if (que == nullptr) {
+        ipc::error("fail: send, queue_of(h) == nullptr\n");
+        return false;
+    }
+    if (que->elems() == nullptr) {
+        ipc::error("fail: send, queue_of(h)->elems() == nullptr\n");
+        return false;
+    }
+    if (!que->ready_sending()) {
+        ipc::error("fail: send, que->ready_sending() == false\n");
+        return false;
+    }
+    ipc::circ::cc_t conns = que->elems()->connections(std::memory_order_relaxed);
+    if (conns == 0) {
+        ipc::error("fail: send, there is no receiver on this connection.\n");
+        return false;
+    }
+    // calc a new message id
+    auto acc = info_of(h)->acc();
+    if (acc == nullptr) {
+        ipc::error("fail: send, info_of(h)->acc() == nullptr\n");
+        return false;
+    }
+    auto msg_id   = acc->fetch_add(1, std::memory_order_relaxed);
+    auto try_push = std::forward<F>(gen_push)(info_of(h), que, msg_id);
+    if (size > ipc::large_msg_limit) {
+        auto   dat = acquire_storage(size, conns);
+        void * buf = dat.second;
+        if (buf != nullptr) {
+            std::memcpy(buf, data, size);
+            return try_push(static_cast<std::int32_t>(size) - 
+                            static_cast<std::int32_t>(ipc::data_length), &(dat.first), 0);
+        }
+        // try using message fragment
+        //ipc::log("fail: shm::handle for big message. msg_id: %zd, size: %zd\n", msg_id, size);
+    }
+    // push message fragment
+    std::int32_t offset = 0;
+    for (std::int32_t i = 0; i < static_cast<std::int32_t>(size / ipc::data_length); ++i, offset += ipc::data_length) {
+        if (!try_push(static_cast<std::int32_t>(size) - offset - static_cast<std::int32_t>(ipc::data_length),
+                      static_cast<ipc::byte_t const *>(data) + offset, ipc::data_length)) {
+            return false;
+        }
+    }
+    // if remain > 0, this is the last message fragment
+    std::int32_t remain = static_cast<std::int32_t>(size) - offset;
+    if (remain > 0) {
+        if (!try_push(remain - static_cast<std::int32_t>(ipc::data_length),
+                      static_cast<ipc::byte_t const *>(data) + offset, 
+                      static_cast<std::size_t>(remain))) {
+            return false;
+        }
+    }
+    return true;
+}
+
+static bool send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
+    return send([tm](auto info, auto que, auto msg_id) {
+        return [tm, info, que, msg_id](std::int32_t remain, void const * data, std::size_t size) {
+            if (!wait_for(info->wt_waiter_, [&] {
+                    return !que->push(
+                        [](void*) { return true; },
+                        info->cc_id_, msg_id, remain, data, size);
+                }, tm)) {
+                ipc::log("force_push: msg_id = %zd, remain = %d, size = %zd\n", msg_id, remain, size);
+                if (!que->force_push(
+                        clear_message<typename queue_t::value_t>,
+                        info->cc_id_, msg_id, remain, data, size)) {
+                    return false;
+                }
+            }
+            info->rd_waiter_.broadcast();
+            return true;
+        };
+    }, h, data, size);
+}
+
+static bool try_send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
+    return send([tm](auto info, auto que, auto msg_id) {
+        return [tm, info, que, msg_id](std::int32_t remain, void const * data, std::size_t size) {
+            if (!wait_for(info->wt_waiter_, [&] {
+                    return !que->push(
+                        [](void*) { return true; },
+                        info->cc_id_, msg_id, remain, data, size);
+                }, tm)) {
+                return false;
+            }
+            info->rd_waiter_.broadcast();
+            return true;
+        };
+    }, h, data, size);
+}
+
+static ipc::buff_t recv(ipc::handle_t h, std::uint64_t tm) {
+    auto que = queue_of(h);
+    if (que == nullptr) {
+        ipc::error("fail: recv, queue_of(h) == nullptr\n");
+        return {};
+    }
+    if (!que->connected()) {
+        // hasn't connected yet, just return.
+        return {};
+    }
+    auto& rc = info_of(h)->recv_cache();
+    for (;;) {
+        // pop a new message
+        typename queue_t::value_t msg;
+        if (!wait_for(info_of(h)->rd_waiter_, [que, &msg] {
+                return !que->pop(msg);
+            }, tm)) {
+            // pop failed, just return.
+            return {};
+        }
+        info_of(h)->wt_waiter_.broadcast();
+        if ((info_of(h)->acc() != nullptr) && (msg.cc_id_ == info_of(h)->cc_id_)) {
+            continue; // ignore message to self
+        }
+        // msg.remain_ may minus & abs(msg.remain_) < data_length
+        std::int32_t r_size = static_cast<std::int32_t>(ipc::data_length) + msg.remain_;
+        if (r_size <= 0) {
+            ipc::error("fail: recv, r_size = %d\n", (int)r_size);
+            return {};
+        }
+        std::size_t msg_size = static_cast<std::size_t>(r_size);
+        // large message
+        if (msg.storage_) {
+            ipc::storage_id_t buf_id = *reinterpret_cast<ipc::storage_id_t*>(&msg.data_);
+            void* buf = find_storage(buf_id, msg_size);
+            if (buf != nullptr) {
+                struct recycle_t {
+                    ipc::storage_id_t storage_id;
+                    ipc::circ::cc_t   curr_conns;
+                    ipc::circ::cc_t   conn_id;
+                } *r_info = ipc::mem::alloc<recycle_t>(recycle_t{
+                    buf_id, que->elems()->connections(std::memory_order_relaxed), que->connected_id()
+                });
+                if (r_info == nullptr) {
+                    ipc::log("fail: ipc::mem::alloc<recycle_t>.\n");
+                    return ipc::buff_t{buf, msg_size}; // no recycle
+                } else {
+                    return ipc::buff_t{buf, msg_size, [](void* p_info, std::size_t size) {
+                        auto r_info = static_cast<recycle_t *>(p_info);
+                        IPC_UNUSED_ auto finally = ipc::guard([r_info] {
+                            ipc::mem::free(r_info);
+                        });
+                        recycle_storage<flag_t>(r_info->storage_id, size, r_info->curr_conns, r_info->conn_id);
+                    }, r_info};
+                }
+            } else {
+                ipc::log("fail: shm::handle for large message. msg_id: %zd, buf_id: %zd, size: %zd\n", msg.id_, buf_id, msg_size);
+                continue;
+            }
+        }
+        // find cache with msg.id_
+        auto cac_it = rc.find(msg.id_);
+        if (cac_it == rc.end()) {
+            if (msg_size <= ipc::data_length) {
+                return make_cache(msg.data_, msg_size);
+            }
+            // gc
+            if (rc.size() > 1024) {
+                std::vector<msg_id_t> need_del;
+                for (auto const & pair : rc) {
+                    auto cmp = std::minmax(msg.id_, pair.first);
+                    if (cmp.second - cmp.first > 8192) {
+                        need_del.push_back(pair.first);
+                    }
+                }
+                for (auto id : need_del) rc.erase(id);
+            }
+            // cache the first message fragment
+            rc.emplace(msg.id_, cache_t { ipc::data_length, make_cache(msg.data_, msg_size) });
+        }
+        // has cached before this message
+        else {
+            auto& cac = cac_it->second;
+            // this is the last message fragment
+            if (msg.remain_ <= 0) {
+                cac.append(&(msg.data_), msg_size);
+                // finish this message, erase it from cache
+                auto buff = std::move(cac.buff_);
+                rc.erase(cac_it);
+                return buff;
+            }
+            // there are remain datas after this message
+            cac.append(&(msg.data_), ipc::data_length);
+        }
+    }
+}
+
+static ipc::buff_t try_recv(ipc::handle_t h) {
+    return recv(h, 0);
+}
+
+}; // detail_impl<Policy>
+
+template <typename Flag>
+using policy_t = ipc::policy::choose<ipc::circ::elem_array, Flag>;
+
+} // internal-linkage
+
+namespace ipc {
+
+template <typename Flag>
+ipc::handle_t chan_impl<Flag>::inited() {
+    ipc::detail::waiter::init();
+    return nullptr;
+}
+
+template <typename Flag>
+bool chan_impl<Flag>::connect(ipc::handle_t * ph, char const * name, unsigned mode) {
+    return detail_impl<policy_t<Flag>>::connect(ph, name, mode & receiver);
+}
+
+template <typename Flag>
+bool chan_impl<Flag>::reconnect(ipc::handle_t * ph, unsigned mode) {
+    return detail_impl<policy_t<Flag>>::reconnect(ph, mode & receiver);
+}
+
+template <typename Flag>
+void chan_impl<Flag>::disconnect(ipc::handle_t h) {
+    detail_impl<policy_t<Flag>>::disconnect(h);
+}
+
+template <typename Flag>
+void chan_impl<Flag>::destroy(ipc::handle_t h) {
+    detail_impl<policy_t<Flag>>::destroy(h);
+}
+
+template <typename Flag>
+char const * chan_impl<Flag>::name(ipc::handle_t h) {
+    auto info = detail_impl<policy_t<Flag>>::info_of(h);
+    return (info == nullptr) ? nullptr : info->name_.c_str();
+}
+
+template <typename Flag>
+std::size_t chan_impl<Flag>::recv_count(ipc::handle_t h) {
+    return detail_impl<policy_t<Flag>>::recv_count(h);
+}
+
+template <typename Flag>
+bool chan_impl<Flag>::wait_for_recv(ipc::handle_t h, std::size_t r_count, std::uint64_t tm) {
+    return detail_impl<policy_t<Flag>>::wait_for_recv(h, r_count, tm);
+}
+
+template <typename Flag>
+bool chan_impl<Flag>::send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
+    return detail_impl<policy_t<Flag>>::send(h, data, size, tm);
+}
+
+template <typename Flag>
+buff_t chan_impl<Flag>::recv(ipc::handle_t h, std::uint64_t tm) {
+    return detail_impl<policy_t<Flag>>::recv(h, tm);
+}
+
+template <typename Flag>
+bool chan_impl<Flag>::try_send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
+    return detail_impl<policy_t<Flag>>::try_send(h, data, size, tm);
+}
+
+template <typename Flag>
+buff_t chan_impl<Flag>::try_recv(ipc::handle_t h) {
+    return detail_impl<policy_t<Flag>>::try_recv(h);
+}
+
+template struct chan_impl<ipc::wr<relat::single, relat::single, trans::unicast  >>;
+// template struct chan_impl<ipc::wr<relat::single, relat::multi , trans::unicast  >>; // TBD
+// template struct chan_impl<ipc::wr<relat::multi , relat::multi , trans::unicast  >>; // TBD
+template struct chan_impl<ipc::wr<relat::single, relat::multi , trans::broadcast>>;
+template struct chan_impl<ipc::wr<relat::multi , relat::multi , trans::broadcast>>;
+
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/policy.h

@@ -0,0 +1,25 @@
+#pragma once
+
+#include <type_traits>
+
+#include "libipc/def.h"
+#include "libipc/prod_cons.h"
+
+#include "libipc/circ/elem_array.h"
+
+namespace ipc {
+namespace policy {
+
+template <template <typename, std::size_t...> class Elems, typename Flag>
+struct choose;
+
+template <typename Flag>
+struct choose<circ::elem_array, Flag> {
+    using flag_t = Flag;
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    using elems_t = circ::elem_array<ipc::prod_cons_impl<flag_t>, DataSize, AlignSize>;
+};
+
+} // namespace policy
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/pool_alloc.cpp

@@ -0,0 +1,17 @@
+#include "libipc/pool_alloc.h"
+
+#include "libipc/memory/resource.h"
+
+namespace ipc {
+namespace mem {
+
+void* pool_alloc::alloc(std::size_t size) {
+    return async_pool_alloc::alloc(size);
+}
+
+void pool_alloc::free(void* p, std::size_t size) {
+    async_pool_alloc::free(p, size);
+}
+
+} // namespace mem
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/prod_cons.h

@@ -0,0 +1,433 @@
+#pragma once
+
+#include <atomic>
+#include <utility>
+#include <cstring>
+#include <type_traits>
+#include <cstdint>
+
+#include "libipc/def.h"
+
+#include "libipc/platform/detail.h"
+#include "libipc/circ/elem_def.h"
+#include "libipc/utility/log.h"
+#include "libipc/utility/utility.h"
+
+namespace ipc {
+
+////////////////////////////////////////////////////////////////
+/// producer-consumer implementation
+////////////////////////////////////////////////////////////////
+
+template <typename Flag>
+struct prod_cons_impl;
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> rd_; // read index
+    alignas(cache_line_size) std::atomic<circ::u2_t> wt_; // write index
+
+    constexpr circ::u2_t cursor() const noexcept {
+        return 0;
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* /*wrapper*/, F&& f, E* elems) {
+        auto cur_wt = circ::index_of(wt_.load(std::memory_order_relaxed));
+        if (cur_wt == circ::index_of(rd_.load(std::memory_order_acquire) - 1)) {
+            return false; // full
+        }
+        std::forward<F>(f)(&(elems[cur_wt].data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    /**
+     * In single-single-unicast, 'force_push' means 'no reader' or 'the only one reader is dead'.
+     * So we could just disconnect all connections of receiver, and return false.
+    */
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(~static_cast<circ::cc_t>(0u));
+        return false;
+    }
+
+    template <typename W, typename F, typename R, typename E>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E* elems) {
+        auto cur_rd = circ::index_of(rd_.load(std::memory_order_relaxed));
+        if (cur_rd == circ::index_of(wt_.load(std::memory_order_acquire))) {
+            return false; // empty
+        }
+        std::forward<F>(f)(&(elems[cur_rd].data_));
+        std::forward<R>(out)(true);
+        rd_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::multi , trans::unicast>>
+     : prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(1);
+        return false;
+    }
+
+    template <typename W, typename F, typename R, 
+              template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
+        byte_t buff[DS];
+        for (unsigned k = 0;;) {
+            auto cur_rd = rd_.load(std::memory_order_relaxed);
+            if (circ::index_of(cur_rd) ==
+                circ::index_of(wt_.load(std::memory_order_acquire))) {
+                return false; // empty
+            }
+            std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
+            if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
+                std::forward<F>(f)(buff);
+                std::forward<R>(out)(true);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::multi , relat::multi, trans::unicast>>
+     : prod_cons_impl<wr<relat::single, relat::multi, trans::unicast>> {
+
+    using flag_t = std::uint64_t;
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<flag_t> f_ct_ { 0 }; // commit flag
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> ct_; // commit index
+
+    template <typename W, typename F, typename E>
+    bool push(W* /*wrapper*/, F&& f, E* elems) {
+        circ::u2_t cur_ct, nxt_ct;
+        for (unsigned k = 0;;) {
+            cur_ct = ct_.load(std::memory_order_relaxed);
+            if (circ::index_of(nxt_ct = cur_ct + 1) ==
+                circ::index_of(rd_.load(std::memory_order_acquire))) {
+                return false; // full
+            }
+            if (ct_.compare_exchange_weak(cur_ct, nxt_ct, std::memory_order_acq_rel)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        auto* el = elems + circ::index_of(cur_ct);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        while (1) {
+            auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
+            if (cur_ct != wt_.load(std::memory_order_relaxed)) {
+                return true;
+            }
+            if ((~cac_ct) != cur_ct) {
+                return true;
+            }
+            if (!el->f_ct_.compare_exchange_strong(cac_ct, 0, std::memory_order_relaxed)) {
+                return true;
+            }
+            wt_.store(nxt_ct, std::memory_order_release);
+            cur_ct = nxt_ct;
+            nxt_ct = cur_ct + 1;
+            el = elems + circ::index_of(cur_ct);
+        }
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(1);
+        return false;
+    }
+
+    template <typename W, typename F, typename R, 
+              template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
+        byte_t buff[DS];
+        for (unsigned k = 0;;) {
+            auto cur_rd = rd_.load(std::memory_order_relaxed);
+            auto cur_wt = wt_.load(std::memory_order_acquire);
+            auto id_rd  = circ::index_of(cur_rd);
+            auto id_wt  = circ::index_of(cur_wt);
+            if (id_rd == id_wt) {
+                auto* el = elems + id_wt;
+                auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
+                if ((~cac_ct) != cur_wt) {
+                    return false; // empty
+                }
+                if (el->f_ct_.compare_exchange_weak(cac_ct, 0, std::memory_order_relaxed)) {
+                    wt_.store(cur_wt + 1, std::memory_order_release);
+                }
+                k = 0;
+            }
+            else {
+                std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
+                if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
+                    std::forward<F>(f)(buff);
+                    std::forward<R>(out)(true);
+                    return true;
+                }
+                ipc::yield(k);
+            }
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::multi, trans::broadcast>> {
+
+    using rc_t = std::uint64_t;
+
+    enum : rc_t {
+        ep_mask = 0x00000000ffffffffull,
+        ep_incr = 0x0000000100000000ull
+    };
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<rc_t> rc_ { 0 }; // read-counter
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> wt_;   // write index
+    alignas(cache_line_size) rc_t epoch_ { 0 };             // only one writer
+
+    circ::u2_t cursor() const noexcept {
+        return wt_.load(std::memory_order_acquire);
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & ep_mask;
+            if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch_)) {
+                return false; // has not finished yet
+            }
+            // consider rem_cc to be 0 here
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        std::forward<F>(f)(&(el->data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        epoch_ += ep_incr;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & ep_mask;
+            if (cc & rem_cc) {
+                ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
+                cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
+                if (cc == 0) return false; // no reader
+            }
+            // just compare & exchange
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        std::forward<F>(f)(&(el->data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename R, typename E>
+    bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E* elems) {
+        if (cur == cursor()) return false; // acquire
+        auto* el = elems + circ::index_of(cur++);
+        std::forward<F>(f)(&(el->data_));
+        for (unsigned k = 0;;) {
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            if ((cur_rc & ep_mask) == 0) {
+                std::forward<R>(out)(true);
+                return true;
+            }
+            auto nxt_rc = cur_rc & ~static_cast<rc_t>(wrapper->connected_id());
+            if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
+                std::forward<R>(out)((nxt_rc & ep_mask) == 0);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::multi, relat::multi, trans::broadcast>> {
+
+    using rc_t   = std::uint64_t;
+    using flag_t = std::uint64_t;
+
+    enum : rc_t {
+        rc_mask = 0x00000000ffffffffull,
+        ep_mask = 0x00ffffffffffffffull,
+        ep_incr = 0x0100000000000000ull,
+        ic_mask = 0xff000000ffffffffull,
+        ic_incr = 0x0000000100000000ull
+    };
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<rc_t  > rc_   { 0 }; // read-counter
+        std::atomic<flag_t> f_ct_ { 0 }; // commit flag
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> ct_;   // commit index
+    alignas(cache_line_size) std::atomic<rc_t> epoch_ { 0 };
+
+    circ::u2_t cursor() const noexcept {
+        return ct_.load(std::memory_order_acquire);
+    }
+
+    constexpr static rc_t inc_rc(rc_t rc) noexcept {
+        return (rc & ic_mask) | ((rc + ic_incr) & ~ic_mask);
+    }
+
+    constexpr static rc_t inc_mask(rc_t rc) noexcept {
+        return inc_rc(rc) & ~rc_mask;
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        circ::u2_t cur_ct;
+        rc_t epoch = epoch_.load(std::memory_order_acquire);
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_relaxed);
+            circ::cc_t rem_cc = cur_rc & rc_mask;
+            if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch)) {
+                return false; // has not finished yet
+            }
+            else if (!rem_cc) {
+                auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
+                if ((cur_fl != cur_ct) && cur_fl) {
+                    return false; // full
+                }
+            }
+            // consider rem_cc to be 0 here
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed) &&
+                epoch_.compare_exchange_weak(epoch, epoch, std::memory_order_acq_rel)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        // only one thread/process would touch here at one time
+        ct_.store(cur_ct + 1, std::memory_order_release);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        circ::u2_t cur_ct;
+        rc_t epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & rc_mask;
+            if (cc & rem_cc) {
+                ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
+                cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
+                if (cc == 0) return false; // no reader
+            }
+            // just compare & exchange
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed)) {
+                if (epoch == epoch_.load(std::memory_order_acquire)) {
+                    break;
+                }
+                else if (push(wrapper, std::forward<F>(f), elems)) {
+                    return true;
+                }
+                epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
+            }
+            ipc::yield(k);
+        }
+        // only one thread/process would touch here at one time
+        ct_.store(cur_ct + 1, std::memory_order_release);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename R, typename E, std::size_t N>
+    bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E(& elems)[N]) {
+        auto* el = elems + circ::index_of(cur);
+        auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
+        if (cur_fl != ~static_cast<flag_t>(cur)) {
+            return false; // empty
+        }
+        ++cur;
+        std::forward<F>(f)(&(el->data_));
+        for (unsigned k = 0;;) {
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            if ((cur_rc & rc_mask) == 0) {
+                std::forward<R>(out)(true);
+                el->f_ct_.store(cur + N - 1, std::memory_order_release);
+                return true;
+            }
+            auto nxt_rc = inc_rc(cur_rc) & ~static_cast<rc_t>(wrapper->connected_id());
+            bool last_one = false;
+            if ((last_one = (nxt_rc & rc_mask) == 0)) {
+                el->f_ct_.store(cur + N - 1, std::memory_order_release);
+            }
+            if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
+                std::forward<R>(out)(last_one);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/queue.h

@@ -0,0 +1,216 @@
+#pragma once
+
+#include <type_traits>
+#include <new>
+#include <utility>  // [[since C++14]]: std::exchange
+#include <algorithm>
+#include <atomic>
+#include <tuple>
+#include <thread>
+#include <chrono>
+#include <string>
+#include <cassert>  // assert
+
+#include "libipc/def.h"
+#include "libipc/shm.h"
+#include "libipc/rw_lock.h"
+
+#include "libipc/utility/log.h"
+#include "libipc/platform/detail.h"
+#include "libipc/circ/elem_def.h"
+
+namespace ipc {
+namespace detail {
+
+class queue_conn {
+protected:
+    circ::cc_t connected_ = 0;
+    shm::handle elems_h_;
+
+    template <typename Elems>
+    Elems* open(char const * name) {
+        if (name == nullptr || name[0] == '\0') {
+            ipc::error("fail open waiter: name is empty!\n");
+            return nullptr;
+        }
+        if (!elems_h_.acquire(name, sizeof(Elems))) {
+            return nullptr;
+        }
+        auto elems = static_cast<Elems*>(elems_h_.get());
+        if (elems == nullptr) {
+            ipc::error("fail acquire elems: %s\n", name);
+            return nullptr;
+        }
+        elems->init();
+        return elems;
+    }
+
+    void close() {
+        elems_h_.release();
+    }
+
+public:
+    queue_conn() = default;
+    queue_conn(const queue_conn&) = delete;
+    queue_conn& operator=(const queue_conn&) = delete;
+
+    bool connected() const noexcept {
+        return connected_ != 0;
+    }
+
+    circ::cc_t connected_id() const noexcept {
+        return connected_;
+    }
+
+    template <typename Elems>
+    auto connect(Elems* elems) noexcept
+                         /*needs 'optional' here*/
+     -> std::tuple<bool, bool, decltype(std::declval<Elems>().cursor())> {
+        if (elems == nullptr) return {};
+        // if it's already connected, just return
+        if (connected()) return {connected(), false, 0};
+        connected_ = elems->connect_receiver();
+        return {connected(), true, elems->cursor()};
+    }
+
+    template <typename Elems>
+    bool disconnect(Elems* elems) noexcept {
+        if (elems == nullptr) return false;
+        // if it's already disconnected, just return false
+        if (!connected()) return false;
+        elems->disconnect_receiver(std::exchange(connected_, 0));
+        return true;
+    }
+};
+
+template <typename Elems>
+class queue_base : public queue_conn {
+    using base_t = queue_conn;
+
+public:
+    using elems_t  = Elems;
+    using policy_t = typename elems_t::policy_t;
+
+protected:
+    elems_t * elems_ = nullptr;
+    decltype(std::declval<elems_t>().cursor()) cursor_ = 0;
+    bool sender_flag_ = false;
+
+public:
+    using base_t::base_t;
+
+    queue_base() = default;
+
+    explicit queue_base(char const * name)
+        : queue_base{} {
+        elems_ = open<elems_t>(name);
+    }
+
+    explicit queue_base(elems_t * elems) noexcept
+        : queue_base{} {
+        assert(elems != nullptr);
+        elems_ = elems;
+    }
+
+    /* not virtual */ ~queue_base() {
+        base_t::close();
+    }
+
+    elems_t       * elems()       noexcept { return elems_; }
+    elems_t const * elems() const noexcept { return elems_; }
+
+    bool ready_sending() noexcept {
+        if (elems_ == nullptr) return false;
+        return sender_flag_ || (sender_flag_ = elems_->connect_sender());
+    }
+
+    void shut_sending() noexcept {
+        if (elems_ == nullptr) return;
+        if (!sender_flag_) return;
+        elems_->disconnect_sender();
+    }
+
+    bool connect() noexcept {
+        auto tp = base_t::connect(elems_);
+        if (std::get<0>(tp) && std::get<1>(tp)) {
+            cursor_ = std::get<2>(tp);
+            return true;
+        }
+        return std::get<0>(tp);
+    }
+
+    bool disconnect() noexcept {
+        return base_t::disconnect(elems_);
+    }
+
+    std::size_t conn_count() const noexcept {
+        return (elems_ == nullptr) ? static_cast<std::size_t>(invalid_value) : elems_->conn_count();
+    }
+
+    bool valid() const noexcept {
+        return elems_ != nullptr;
+    }
+
+    bool empty() const noexcept {
+        return !valid() || (cursor_ == elems_->cursor());
+    }
+
+    template <typename T, typename F, typename... P>
+    bool push(F&& prep, P&&... params) {
+        if (elems_ == nullptr) return false;
+        return elems_->push(this, [&](void* p) {
+            if (prep(p)) ::new (p) T(std::forward<P>(params)...);
+        });
+    }
+
+    template <typename T, typename F, typename... P>
+    bool force_push(F&& prep, P&&... params) {
+        if (elems_ == nullptr) return false;
+        return elems_->force_push(this, [&](void* p) {
+            if (prep(p)) ::new (p) T(std::forward<P>(params)...);
+        });
+    }
+
+    template <typename T, typename F>
+    bool pop(T& item, F&& out) {
+        if (elems_ == nullptr) {
+            return false;
+        }
+        return elems_->pop(this, &(this->cursor_), [&item](void* p) {
+            ::new (&item) T(std::move(*static_cast<T*>(p)));
+        }, std::forward<F>(out));
+    }
+};
+
+} // namespace detail
+
+template <typename T, typename Policy>
+class queue final : public detail::queue_base<typename Policy::template elems_t<sizeof(T), alignof(T)>> {
+    using base_t = detail::queue_base<typename Policy::template elems_t<sizeof(T), alignof(T)>>;
+
+public:
+    using value_t = T;
+
+    using base_t::base_t;
+
+    template <typename... P>
+    bool push(P&&... params) {
+        return base_t::template push<T>(std::forward<P>(params)...);
+    }
+
+    template <typename... P>
+    bool force_push(P&&... params) {
+        return base_t::template force_push<T>(std::forward<P>(params)...);
+    }
+
+    bool pop(T& item) {
+        return base_t::pop(item, [](bool) {});
+    }
+
+    template <typename F>
+    bool pop(T& item, F&& out) {
+        return base_t::pop(item, std::forward<F>(out));
+    }
+};
+
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/shm.cpp

@@ -0,0 +1,103 @@
+
+#include <string>
+#include <utility>
+
+#include "libipc/shm.h"
+
+#include "libipc/utility/pimpl.h"
+#include "libipc/memory/resource.h"
+
+namespace ipc {
+namespace shm {
+
+class handle::handle_ : public pimpl<handle_> {
+public:
+    shm::id_t id_ = nullptr;
+    void*     m_  = nullptr;
+
+    ipc::string n_;
+    std::size_t s_ = 0;
+};
+
+handle::handle()
+    : p_(p_->make()) {
+}
+
+handle::handle(char const * name, std::size_t size, unsigned mode)
+    : handle() {
+    acquire(name, size, mode);
+}
+
+handle::handle(handle&& rhs)
+    : handle() {
+    swap(rhs);
+}
+
+handle::~handle() {
+    release();
+    p_->clear();
+}
+
+void handle::swap(handle& rhs) {
+    std::swap(p_, rhs.p_);
+}
+
+handle& handle::operator=(handle rhs) {
+    swap(rhs);
+    return *this;
+}
+
+bool handle::valid() const noexcept {
+    return impl(p_)->m_ != nullptr;
+}
+
+std::size_t handle::size() const noexcept {
+    return impl(p_)->s_;
+}
+
+char const * handle::name() const noexcept {
+    return impl(p_)->n_.c_str();
+}
+
+std::int32_t handle::ref() const noexcept {
+    return shm::get_ref(impl(p_)->id_);
+}
+
+void handle::sub_ref() noexcept {
+    shm::sub_ref(impl(p_)->id_);
+}
+
+bool handle::acquire(char const * name, std::size_t size, unsigned mode) {
+    release();
+    impl(p_)->id_ = shm::acquire((impl(p_)->n_ = name).c_str(), size, mode);
+    impl(p_)->m_  = shm::get_mem(impl(p_)->id_, &(impl(p_)->s_));
+    return valid();
+}
+
+std::int32_t handle::release() {
+    if (impl(p_)->id_ == nullptr) return -1;
+    return shm::release(detach());
+}
+
+void* handle::get() const {
+    return impl(p_)->m_;
+}
+
+void handle::attach(id_t id) {
+    if (id == nullptr) return;
+    release();
+    impl(p_)->id_ = id;
+    impl(p_)->m_  = shm::get_mem(impl(p_)->id_, &(impl(p_)->s_));
+}
+
+id_t handle::detach() {
+    auto old = impl(p_)->id_;
+    impl(p_)->id_ = nullptr;
+    impl(p_)->m_  = nullptr;
+    impl(p_)->s_  = 0;
+    impl(p_)->n_.clear();
+    return old;
+}
+
+} // namespace shm
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/waiter.h

@@ -0,0 +1,83 @@
+#pragma once
+
+#include <utility>
+#include <string>
+#include <mutex>
+#include <atomic>
+
+#include "libipc/def.h"
+#include "libipc/mutex.h"
+#include "libipc/condition.h"
+#include "libipc/platform/detail.h"
+
+namespace ipc {
+namespace detail {
+
+class waiter {
+    ipc::sync::condition cond_;
+    ipc::sync::mutex     lock_;
+    std::atomic<bool>    quit_ {false};
+
+public:
+    static void init();
+
+    waiter() = default;
+    waiter(char const *name) {
+        open(name);
+    }
+
+    ~waiter() {
+        close();
+    }
+
+    bool valid() const noexcept {
+        return cond_.valid() && lock_.valid();
+    }
+
+    bool open(char const *name) noexcept {
+        quit_.store(false, std::memory_order_relaxed);
+        if (!cond_.open((std::string{"_waiter_cond_"} + name).c_str())) {
+            return false;
+        }
+        if (!lock_.open((std::string{"_waiter_lock_"} + name).c_str())) {
+            cond_.close();
+            return false;
+        }
+        return valid();
+    }
+
+    void close() noexcept {
+        cond_.close();
+        lock_.close();
+    }
+
+    template <typename F>
+    bool wait_if(F &&pred, std::uint64_t tm = ipc::invalid_value) noexcept {
+        IPC_UNUSED_ std::lock_guard<ipc::sync::mutex> guard {lock_};
+        while ([this, &pred] {
+                    return !quit_.load(std::memory_order_relaxed)
+                        && std::forward<F>(pred)();
+                }()) {
+            if (!cond_.wait(lock_, tm)) return false;
+        }
+        return true;
+    }
+
+    bool notify() noexcept {
+        std::lock_guard<ipc::sync::mutex>{lock_}; // barrier
+        return cond_.notify(lock_);
+    }
+
+    bool broadcast() noexcept {
+        std::lock_guard<ipc::sync::mutex>{lock_}; // barrier
+        return cond_.broadcast(lock_);
+    }
+
+    bool quit_waiting() {
+        quit_.store(true, std::memory_order_release);
+        return broadcast();
+    }
+};
+
+} // namespace detail
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/来源

@@ -0,0 +1,3 @@
+https://github.com/mutouyun/cpp-ipc
+
+A high-performance inter-process communication library using shared memory on Linux/Windows.

crazy_functions/test_project/cpp/libJPG/jpgd.h

@@ -0,0 +1,316 @@
+// jpgd.h - C++ class for JPEG decompression.
+// Public domain, Rich Geldreich <richgel99@gmail.com>
+#ifndef JPEG_DECODER_H
+#define JPEG_DECODER_H
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <setjmp.h>
+
+namespace jpgd
+{
+  typedef unsigned char  uint8;
+  typedef   signed short int16;
+  typedef unsigned short uint16;
+  typedef unsigned int   uint;
+  typedef   signed int   int32;
+
+  // Loads a JPEG image from a memory buffer or a file.
+  // req_comps can be 1 (grayscale), 3 (RGB), or 4 (RGBA).
+  // On return, width/height will be set to the image's dimensions, and actual_comps will be set to the either 1 (grayscale) or 3 (RGB).
+  // Notes: For more control over where and how the source data is read, see the decompress_jpeg_image_from_stream() function below, or call the jpeg_decoder class directly.
+  // Requesting a 8 or 32bpp image is currently a little faster than 24bpp because the jpeg_decoder class itself currently always unpacks to either 8 or 32bpp.
+// BEGIN EPIC MOD
+//unsigned char *decompress_jpeg_image_from_memory(const unsigned char *pSrc_data, int src_data_size, int *width, int *height, int *actual_comps, int req_comps);
+  unsigned char *decompress_jpeg_image_from_memory(const unsigned char *pSrc_data, int src_data_size, int *width, int *height, int *actual_comps, int req_comps, int format);
+// END EPIC MOD
+  unsigned char *decompress_jpeg_image_from_file(const char *pSrc_filename, int *width, int *height, int *actual_comps, int req_comps);
+
+  // Success/failure error codes.
+  enum jpgd_status
+  {
+    JPGD_SUCCESS = 0, JPGD_FAILED = -1, JPGD_DONE = 1,
+    JPGD_BAD_DHT_COUNTS = -256, JPGD_BAD_DHT_INDEX, JPGD_BAD_DHT_MARKER, JPGD_BAD_DQT_MARKER, JPGD_BAD_DQT_TABLE, 
+    JPGD_BAD_PRECISION, JPGD_BAD_HEIGHT, JPGD_BAD_WIDTH, JPGD_TOO_MANY_COMPONENTS, 
+    JPGD_BAD_SOF_LENGTH, JPGD_BAD_VARIABLE_MARKER, JPGD_BAD_DRI_LENGTH, JPGD_BAD_SOS_LENGTH,
+    JPGD_BAD_SOS_COMP_ID, JPGD_W_EXTRA_BYTES_BEFORE_MARKER, JPGD_NO_ARITHMITIC_SUPPORT, JPGD_UNEXPECTED_MARKER,
+    JPGD_NOT_JPEG, JPGD_UNSUPPORTED_MARKER, JPGD_BAD_DQT_LENGTH, JPGD_TOO_MANY_BLOCKS,
+    JPGD_UNDEFINED_QUANT_TABLE, JPGD_UNDEFINED_HUFF_TABLE, JPGD_NOT_SINGLE_SCAN, JPGD_UNSUPPORTED_COLORSPACE,
+    JPGD_UNSUPPORTED_SAMP_FACTORS, JPGD_DECODE_ERROR, JPGD_BAD_RESTART_MARKER, JPGD_ASSERTION_ERROR,
+    JPGD_BAD_SOS_SPECTRAL, JPGD_BAD_SOS_SUCCESSIVE, JPGD_STREAM_READ, JPGD_NOTENOUGHMEM
+  };
+    
+  // Input stream interface.
+  // Derive from this class to read input data from sources other than files or memory. Set m_eof_flag to true when no more data is available.
+  // The decoder is rather greedy: it will keep on calling this method until its internal input buffer is full, or until the EOF flag is set.
+  // It the input stream contains data after the JPEG stream's EOI (end of image) marker it will probably be pulled into the internal buffer.
+  // Call the get_total_bytes_read() method to determine the actual size of the JPEG stream after successful decoding.
+  class jpeg_decoder_stream
+  {
+  public:
+    jpeg_decoder_stream() { }
+    virtual ~jpeg_decoder_stream() { }
+
+    // The read() method is called when the internal input buffer is empty.
+    // Parameters:
+    // pBuf - input buffer
+    // max_bytes_to_read - maximum bytes that can be written to pBuf
+    // pEOF_flag - set this to true if at end of stream (no more bytes remaining)
+    // Returns -1 on error, otherwise return the number of bytes actually written to the buffer (which may be 0).
+    // Notes: This method will be called in a loop until you set *pEOF_flag to true or the internal buffer is full.
+    virtual int read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag) = 0;
+  };
+
+  // stdio FILE stream class.
+  class jpeg_decoder_file_stream : public jpeg_decoder_stream
+  {
+    jpeg_decoder_file_stream(const jpeg_decoder_file_stream &);
+    jpeg_decoder_file_stream &operator =(const jpeg_decoder_file_stream &);
+
+    FILE *m_pFile;
+    bool m_eof_flag, m_error_flag;
+
+  public:
+    jpeg_decoder_file_stream();
+    virtual ~jpeg_decoder_file_stream();
+    
+    bool open(const char *Pfilename);
+    void close();
+
+    virtual int read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag);
+  };
+
+  // Memory stream class.
+  class jpeg_decoder_mem_stream : public jpeg_decoder_stream
+  {
+    const uint8 *m_pSrc_data;
+    uint m_ofs, m_size;
+
+  public:
+    jpeg_decoder_mem_stream() : m_pSrc_data(NULL), m_ofs(0), m_size(0) { }
+    jpeg_decoder_mem_stream(const uint8 *pSrc_data, uint size) : m_pSrc_data(pSrc_data), m_ofs(0), m_size(size) { }
+
+    virtual ~jpeg_decoder_mem_stream() { }
+
+    bool open(const uint8 *pSrc_data, uint size);
+    void close() { m_pSrc_data = NULL; m_ofs = 0; m_size = 0; }
+    
+    virtual int read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag);
+  };
+
+  // Loads JPEG file from a jpeg_decoder_stream.
+  unsigned char *decompress_jpeg_image_from_stream(jpeg_decoder_stream *pStream, int *width, int *height, int *actual_comps, int req_comps);
+
+  enum 
+  { 
+    JPGD_IN_BUF_SIZE = 8192, JPGD_MAX_BLOCKS_PER_MCU = 10, JPGD_MAX_HUFF_TABLES = 8, JPGD_MAX_QUANT_TABLES = 4, 
+    JPGD_MAX_COMPONENTS = 4, JPGD_MAX_COMPS_IN_SCAN = 4, JPGD_MAX_BLOCKS_PER_ROW = 8192, JPGD_MAX_HEIGHT = 16384, JPGD_MAX_WIDTH = 16384 
+  };
+          
+  typedef int16 jpgd_quant_t;
+  typedef int16 jpgd_block_t;
+
+  class jpeg_decoder
+  {
+  public:
+    // Call get_error_code() after constructing to determine if the stream is valid or not. You may call the get_width(), get_height(), etc.
+    // methods after the constructor is called. You may then either destruct the object, or begin decoding the image by calling begin_decoding(), then decode() on each scanline.
+    jpeg_decoder(jpeg_decoder_stream *pStream);
+
+    ~jpeg_decoder();
+
+    // Call this method after constructing the object to begin decompression.
+    // If JPGD_SUCCESS is returned you may then call decode() on each scanline.
+    int begin_decoding();
+
+    // Returns the next scan line.
+    // For grayscale images, pScan_line will point to a buffer containing 8-bit pixels (get_bytes_per_pixel() will return 1). 
+    // Otherwise, it will always point to a buffer containing 32-bit RGBA pixels (A will always be 255, and get_bytes_per_pixel() will return 4).
+    // Returns JPGD_SUCCESS if a scan line has been returned.
+    // Returns JPGD_DONE if all scan lines have been returned.
+    // Returns JPGD_FAILED if an error occurred. Call get_error_code() for a more info.
+    int decode(const void** pScan_line, uint* pScan_line_len);
+    
+    inline jpgd_status get_error_code() const { return m_error_code; }
+
+    inline int get_width() const { return m_image_x_size; }
+    inline int get_height() const { return m_image_y_size; }
+
+    inline int get_num_components() const { return m_comps_in_frame; }
+
+    inline int get_bytes_per_pixel() const { return m_dest_bytes_per_pixel; }
+    inline int get_bytes_per_scan_line() const { return m_image_x_size * get_bytes_per_pixel(); }
+
+    // Returns the total number of bytes actually consumed by the decoder (which should equal the actual size of the JPEG file).
+    inline int get_total_bytes_read() const { return m_total_bytes_read; }
+    
+  private:
+    jpeg_decoder(const jpeg_decoder &);
+    jpeg_decoder &operator =(const jpeg_decoder &);
+
+    typedef void (*pDecode_block_func)(jpeg_decoder *, int, int, int);
+
+    struct huff_tables
+    {
+      bool ac_table;
+      uint  look_up[256];
+      uint  look_up2[256];
+      uint8 code_size[256];
+      uint  tree[512];
+    };
+
+    struct coeff_buf
+    {
+      uint8 *pData;
+      int block_num_x, block_num_y;
+      int block_len_x, block_len_y;
+      int block_size;
+    };
+
+    struct mem_block
+    {
+      mem_block *m_pNext;
+      size_t m_used_count;
+      size_t m_size;
+      char m_data[1];
+    };
+
+    jmp_buf m_jmp_state;
+    mem_block *m_pMem_blocks;
+    int m_image_x_size;
+    int m_image_y_size;
+    jpeg_decoder_stream *m_pStream;
+    int m_progressive_flag;
+    uint8 m_huff_ac[JPGD_MAX_HUFF_TABLES];
+    uint8* m_huff_num[JPGD_MAX_HUFF_TABLES];      // pointer to number of Huffman codes per bit size
+    uint8* m_huff_val[JPGD_MAX_HUFF_TABLES];      // pointer to Huffman codes per bit size
+    jpgd_quant_t* m_quant[JPGD_MAX_QUANT_TABLES]; // pointer to quantization tables
+    int m_scan_type;                              // Gray, Yh1v1, Yh1v2, Yh2v1, Yh2v2 (CMYK111, CMYK4114 no longer supported)
+    int m_comps_in_frame;                         // # of components in frame
+    int m_comp_h_samp[JPGD_MAX_COMPONENTS];       // component's horizontal sampling factor
+    int m_comp_v_samp[JPGD_MAX_COMPONENTS];       // component's vertical sampling factor
+    int m_comp_quant[JPGD_MAX_COMPONENTS];        // component's quantization table selector
+    int m_comp_ident[JPGD_MAX_COMPONENTS];        // component's ID
+    int m_comp_h_blocks[JPGD_MAX_COMPONENTS];
+    int m_comp_v_blocks[JPGD_MAX_COMPONENTS];
+    int m_comps_in_scan;                          // # of components in scan
+    int m_comp_list[JPGD_MAX_COMPS_IN_SCAN];      // components in this scan
+    int m_comp_dc_tab[JPGD_MAX_COMPONENTS];       // component's DC Huffman coding table selector
+    int m_comp_ac_tab[JPGD_MAX_COMPONENTS];       // component's AC Huffman coding table selector
+    int m_spectral_start;                         // spectral selection start
+    int m_spectral_end;                           // spectral selection end
+    int m_successive_low;                         // successive approximation low
+    int m_successive_high;                        // successive approximation high
+    int m_max_mcu_x_size;                         // MCU's max. X size in pixels
+    int m_max_mcu_y_size;                         // MCU's max. Y size in pixels
+    int m_blocks_per_mcu;
+    int m_max_blocks_per_row;
+    int m_mcus_per_row, m_mcus_per_col;
+    int m_mcu_org[JPGD_MAX_BLOCKS_PER_MCU];
+    int m_total_lines_left;                       // total # lines left in image
+    int m_mcu_lines_left;                         // total # lines left in this MCU
+    int m_real_dest_bytes_per_scan_line;
+    int m_dest_bytes_per_scan_line;               // rounded up
+    int m_dest_bytes_per_pixel;                   // 4 (RGB) or 1 (Y)
+    huff_tables* m_pHuff_tabs[JPGD_MAX_HUFF_TABLES];
+    coeff_buf* m_dc_coeffs[JPGD_MAX_COMPONENTS];
+    coeff_buf* m_ac_coeffs[JPGD_MAX_COMPONENTS];
+    int m_eob_run;
+    int m_block_y_mcu[JPGD_MAX_COMPONENTS];
+    uint8* m_pIn_buf_ofs;
+    int m_in_buf_left;
+    int m_tem_flag;
+    bool m_eof_flag;
+    uint8 m_in_buf_pad_start[128];
+    uint8 m_in_buf[JPGD_IN_BUF_SIZE + 128];
+    uint8 m_in_buf_pad_end[128];
+    int m_bits_left;
+    uint m_bit_buf;
+    int m_restart_interval;
+    int m_restarts_left;
+    int m_next_restart_num;
+    int m_max_mcus_per_row;
+    int m_max_blocks_per_mcu;
+    int m_expanded_blocks_per_mcu;
+    int m_expanded_blocks_per_row;
+    int m_expanded_blocks_per_component;
+    bool  m_freq_domain_chroma_upsample;
+    int m_max_mcus_per_col;
+    uint m_last_dc_val[JPGD_MAX_COMPONENTS];
+    jpgd_block_t* m_pMCU_coefficients;
+    int m_mcu_block_max_zag[JPGD_MAX_BLOCKS_PER_MCU];
+    uint8* m_pSample_buf;
+    int m_crr[256];
+    int m_cbb[256];
+    int m_crg[256];
+    int m_cbg[256];
+    uint8* m_pScan_line_0;
+    uint8* m_pScan_line_1;
+    jpgd_status m_error_code;
+    bool m_ready_flag;
+    int m_total_bytes_read;
+
+    void free_all_blocks();
+    // BEGIN EPIC MOD
+    UE_NORETURN void stop_decoding(jpgd_status status);
+    // END EPIC MOD
+    void *alloc(size_t n, bool zero = false);
+    void word_clear(void *p, uint16 c, uint n);
+    void prep_in_buffer();
+    void read_dht_marker();
+    void read_dqt_marker();
+    void read_sof_marker();
+    void skip_variable_marker();
+    void read_dri_marker();
+    void read_sos_marker();
+    int next_marker();
+    int process_markers();
+    void locate_soi_marker();
+    void locate_sof_marker();
+    int locate_sos_marker();
+    void init(jpeg_decoder_stream * pStream);
+    void create_look_ups();
+    void fix_in_buffer();
+    void transform_mcu(int mcu_row);
+    void transform_mcu_expand(int mcu_row);
+    coeff_buf* coeff_buf_open(int block_num_x, int block_num_y, int block_len_x, int block_len_y);
+    inline jpgd_block_t *coeff_buf_getp(coeff_buf *cb, int block_x, int block_y);
+    void load_next_row();
+    void decode_next_row();
+    void make_huff_table(int index, huff_tables *pH);
+    void check_quant_tables();
+    void check_huff_tables();
+    void calc_mcu_block_order();
+    int init_scan();
+    void init_frame();
+    void process_restart();
+    void decode_scan(pDecode_block_func decode_block_func);
+    void init_progressive();
+    void init_sequential();
+    void decode_start();
+    void decode_init(jpeg_decoder_stream * pStream);
+    void H2V2Convert();
+    void H2V1Convert();
+    void H1V2Convert();
+    void H1V1Convert();
+    void gray_convert();
+    void expanded_convert();
+    void find_eoi();
+    inline uint get_char();
+    inline uint get_char(bool *pPadding_flag);
+    inline void stuff_char(uint8 q);
+    inline uint8 get_octet();
+    inline uint get_bits(int num_bits);
+    inline uint get_bits_no_markers(int numbits);
+    inline int huff_decode(huff_tables *pH);
+    inline int huff_decode(huff_tables *pH, int& extrabits);
+    static inline uint8 clamp(int i);
+    static void decode_block_dc_first(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+    static void decode_block_dc_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+    static void decode_block_ac_first(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+    static void decode_block_ac_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+  };
+  
+} // namespace jpgd
+
+#endif // JPEG_DECODER_H

crazy_functions/test_project/cpp/libJPG/jpge.h

@@ -0,0 +1,172 @@
+
+// jpge.h - C++ class for JPEG compression.
+// Public domain, Rich Geldreich <richgel99@gmail.com>
+// Alex Evans: Added RGBA support, linear memory allocator.
+#ifndef JPEG_ENCODER_H
+#define JPEG_ENCODER_H
+
+#include <stdint.h>
+
+namespace jpge
+{
+  typedef unsigned char  uint8;
+  typedef signed short   int16;
+  typedef signed int     int32;
+  typedef unsigned short uint16;
+  typedef unsigned int   uint32;
+  typedef unsigned int   uint;
+
+  // JPEG chroma subsampling factors. Y_ONLY (grayscale images) and H2V2 (color images) are the most common.
+  enum subsampling_t { Y_ONLY = 0, H1V1 = 1, H2V1 = 2, H2V2 = 3 };
+
+  // JPEG compression parameters structure.
+  struct params
+  {
+    inline params() : m_quality(85), m_subsampling(H2V2), m_no_chroma_discrim_flag(false), m_two_pass_flag(false) { }
+
+    inline bool check_valid() const
+    {
+      if ((m_quality < 1) || (m_quality > 100)) return false;
+      if ((uint)m_subsampling > (uint)H2V2) return false;
+      return true;
+    }
+
+    // Quality: 1-100, higher is better. Typical values are around 50-95.
+    int m_quality;
+
+    // m_subsampling:
+    // 0 = Y (grayscale) only
+    // 1 = YCbCr, no subsampling (H1V1, YCbCr 1x1x1, 3 blocks per MCU)
+    // 2 = YCbCr, H2V1 subsampling (YCbCr 2x1x1, 4 blocks per MCU)
+    // 3 = YCbCr, H2V2 subsampling (YCbCr 4x1x1, 6 blocks per MCU-- very common)
+    subsampling_t m_subsampling;
+
+    // Disables CbCr discrimination - only intended for testing.
+    // If true, the Y quantization table is also used for the CbCr channels.
+    bool m_no_chroma_discrim_flag;
+
+    bool m_two_pass_flag;
+  };
+  
+  // Writes JPEG image to a file. 
+  // num_channels must be 1 (Y) or 3 (RGB), image pitch must be width*num_channels.
+  bool compress_image_to_jpeg_file(const char *pFilename, int64_t width, int64_t height, int64_t num_channels, const uint8 *pImage_data, const params &comp_params = params());
+
+  // Writes JPEG image to memory buffer. 
+  // On entry, buf_size is the size of the output buffer pointed at by pBuf, which should be at least ~1024 bytes. 
+  // If return value is true, buf_size will be set to the size of the compressed data.
+  bool compress_image_to_jpeg_file_in_memory(void *pBuf, int64_t &buf_size, int64_t width, int64_t height, int64_t num_channels, const uint8 *pImage_data, const params &comp_params = params());
+    
+  // Output stream abstract class - used by the jpeg_encoder class to write to the output stream. 
+  // put_buf() is generally called with len==JPGE_OUT_BUF_SIZE bytes, but for headers it'll be called with smaller amounts.
+  class output_stream
+  {
+  public:
+    virtual ~output_stream() { };
+    virtual bool put_buf(const void* Pbuf, int64_t len) = 0;
+    template<class T> inline bool put_obj(const T& obj) { return put_buf(&obj, sizeof(T)); }
+  };
+    
+  // Lower level jpeg_encoder class - useful if more control is needed than the above helper functions.
+  class jpeg_encoder
+  {
+  public:
+    jpeg_encoder();
+    ~jpeg_encoder();
+
+    // Initializes the compressor.
+    // pStream: The stream object to use for writing compressed data.
+    // params - Compression parameters structure, defined above.
+    // width, height  - Image dimensions.
+    // channels - May be 1, or 3. 1 indicates grayscale, 3 indicates RGB source data.
+    // Returns false on out of memory or if a stream write fails.
+    bool init(output_stream *pStream, int64_t width, int64_t height, int64_t src_channels, const params &comp_params = params());
+    
+    const params &get_params() const { return m_params; }
+    
+    // Deinitializes the compressor, freeing any allocated memory. May be called at any time.
+    void deinit();
+
+    uint get_total_passes() const { return m_params.m_two_pass_flag ? 2 : 1; }
+    inline uint get_cur_pass() { return m_pass_num; }
+
+    // Call this method with each source scanline.
+    // width * src_channels bytes per scanline is expected (RGB or Y format).
+    // You must call with NULL after all scanlines are processed to finish compression.
+    // Returns false on out of memory or if a stream write fails.
+    bool process_scanline(const void* pScanline);
+        
+  private:
+    jpeg_encoder(const jpeg_encoder &);
+    jpeg_encoder &operator =(const jpeg_encoder &);
+
+    typedef int32 sample_array_t;
+        
+    output_stream *m_pStream;
+    params m_params;
+    uint8 m_num_components;
+    uint8 m_comp_h_samp[3], m_comp_v_samp[3];
+    int m_image_x, m_image_y, m_image_bpp, m_image_bpl;
+    int m_image_x_mcu, m_image_y_mcu;
+    int m_image_bpl_xlt, m_image_bpl_mcu;
+    int m_mcus_per_row;
+    int m_mcu_x, m_mcu_y;
+    uint8 *m_mcu_lines[16];
+    uint8 m_mcu_y_ofs;
+    sample_array_t m_sample_array[64];
+    int16 m_coefficient_array[64];
+    int32 m_quantization_tables[2][64];
+    uint m_huff_codes[4][256];
+    uint8 m_huff_code_sizes[4][256];
+    uint8 m_huff_bits[4][17];
+    uint8 m_huff_val[4][256];
+    uint32 m_huff_count[4][256];
+    int m_last_dc_val[3];
+    enum { JPGE_OUT_BUF_SIZE = 2048 };
+    uint8 m_out_buf[JPGE_OUT_BUF_SIZE];
+    uint8 *m_pOut_buf;
+    uint m_out_buf_left;
+    uint32 m_bit_buffer;
+    uint m_bits_in;
+    uint8 m_pass_num;
+    bool m_all_stream_writes_succeeded;
+        
+    void optimize_huffman_table(int table_num, int table_len);
+    void emit_byte(uint8 i);
+    void emit_word(uint i);
+    void emit_marker(int marker);
+    void emit_jfif_app0();
+    void emit_dqt();
+    void emit_sof();
+    void emit_dht(uint8 *bits, uint8 *val, int index, bool ac_flag);
+    void emit_dhts();
+    void emit_sos();
+    void emit_markers();
+    void compute_huffman_table(uint *codes, uint8 *code_sizes, uint8 *bits, uint8 *val);
+    void compute_quant_table(int32 *dst, int16 *src);
+    void adjust_quant_table(int32 *dst, int32 *src);
+    void first_pass_init();
+    bool second_pass_init();
+    bool jpg_open(int p_x_res, int p_y_res, int src_channels);
+    void load_block_8_8_grey(int x);
+    void load_block_8_8(int x, int y, int c);
+    void load_block_16_8(int x, int c);
+    void load_block_16_8_8(int x, int c);
+    void load_quantized_coefficients(int component_num);
+    void flush_output_buffer();
+    void put_bits(uint bits, uint len);
+    void code_coefficients_pass_one(int component_num);
+    void code_coefficients_pass_two(int component_num);
+    void code_block(int component_num);
+    void process_mcu_row();
+    bool terminate_pass_one();
+    bool terminate_pass_two();
+    bool process_end_of_image();
+    void load_mcu(const void* src);
+    void clear();
+    void init();
+  };
+
+} // namespace jpge
+
+#endif // JPEG_ENCODER

crazy_functions/test_project/cpp/libJPG/来源

@@ -0,0 +1,3 @@
+jpge.h - C++ class for JPEG compression.
+Public domain, Rich Geldreich <richgel99@gmail.com>
+Alex Evans: Added RGBA support, linear memory allocator.

crazy_functions/test_project/cpp/longcode/prod_cons.h

@@ -0,0 +1,433 @@
+#pragma once
+
+#include <atomic>
+#include <utility>
+#include <cstring>
+#include <type_traits>
+#include <cstdint>
+
+#include "libipc/def.h"
+
+#include "libipc/platform/detail.h"
+#include "libipc/circ/elem_def.h"
+#include "libipc/utility/log.h"
+#include "libipc/utility/utility.h"
+
+namespace ipc {
+
+////////////////////////////////////////////////////////////////
+/// producer-consumer implementation
+////////////////////////////////////////////////////////////////
+
+template <typename Flag>
+struct prod_cons_impl;
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> rd_; // read index
+    alignas(cache_line_size) std::atomic<circ::u2_t> wt_; // write index
+
+    constexpr circ::u2_t cursor() const noexcept {
+        return 0;
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* /*wrapper*/, F&& f, E* elems) {
+        auto cur_wt = circ::index_of(wt_.load(std::memory_order_relaxed));
+        if (cur_wt == circ::index_of(rd_.load(std::memory_order_acquire) - 1)) {
+            return false; // full
+        }
+        std::forward<F>(f)(&(elems[cur_wt].data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    /**
+     * In single-single-unicast, 'force_push' means 'no reader' or 'the only one reader is dead'.
+     * So we could just disconnect all connections of receiver, and return false.
+    */
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(~static_cast<circ::cc_t>(0u));
+        return false;
+    }
+
+    template <typename W, typename F, typename R, typename E>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E* elems) {
+        auto cur_rd = circ::index_of(rd_.load(std::memory_order_relaxed));
+        if (cur_rd == circ::index_of(wt_.load(std::memory_order_acquire))) {
+            return false; // empty
+        }
+        std::forward<F>(f)(&(elems[cur_rd].data_));
+        std::forward<R>(out)(true);
+        rd_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::multi , trans::unicast>>
+     : prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(1);
+        return false;
+    }
+
+    template <typename W, typename F, typename R, 
+              template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
+        byte_t buff[DS];
+        for (unsigned k = 0;;) {
+            auto cur_rd = rd_.load(std::memory_order_relaxed);
+            if (circ::index_of(cur_rd) ==
+                circ::index_of(wt_.load(std::memory_order_acquire))) {
+                return false; // empty
+            }
+            std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
+            if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
+                std::forward<F>(f)(buff);
+                std::forward<R>(out)(true);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::multi , relat::multi, trans::unicast>>
+     : prod_cons_impl<wr<relat::single, relat::multi, trans::unicast>> {
+
+    using flag_t = std::uint64_t;
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<flag_t> f_ct_ { 0 }; // commit flag
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> ct_; // commit index
+
+    template <typename W, typename F, typename E>
+    bool push(W* /*wrapper*/, F&& f, E* elems) {
+        circ::u2_t cur_ct, nxt_ct;
+        for (unsigned k = 0;;) {
+            cur_ct = ct_.load(std::memory_order_relaxed);
+            if (circ::index_of(nxt_ct = cur_ct + 1) ==
+                circ::index_of(rd_.load(std::memory_order_acquire))) {
+                return false; // full
+            }
+            if (ct_.compare_exchange_weak(cur_ct, nxt_ct, std::memory_order_acq_rel)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        auto* el = elems + circ::index_of(cur_ct);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        while (1) {
+            auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
+            if (cur_ct != wt_.load(std::memory_order_relaxed)) {
+                return true;
+            }
+            if ((~cac_ct) != cur_ct) {
+                return true;
+            }
+            if (!el->f_ct_.compare_exchange_strong(cac_ct, 0, std::memory_order_relaxed)) {
+                return true;
+            }
+            wt_.store(nxt_ct, std::memory_order_release);
+            cur_ct = nxt_ct;
+            nxt_ct = cur_ct + 1;
+            el = elems + circ::index_of(cur_ct);
+        }
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(1);
+        return false;
+    }
+
+    template <typename W, typename F, typename R, 
+              template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
+        byte_t buff[DS];
+        for (unsigned k = 0;;) {
+            auto cur_rd = rd_.load(std::memory_order_relaxed);
+            auto cur_wt = wt_.load(std::memory_order_acquire);
+            auto id_rd  = circ::index_of(cur_rd);
+            auto id_wt  = circ::index_of(cur_wt);
+            if (id_rd == id_wt) {
+                auto* el = elems + id_wt;
+                auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
+                if ((~cac_ct) != cur_wt) {
+                    return false; // empty
+                }
+                if (el->f_ct_.compare_exchange_weak(cac_ct, 0, std::memory_order_relaxed)) {
+                    wt_.store(cur_wt + 1, std::memory_order_release);
+                }
+                k = 0;
+            }
+            else {
+                std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
+                if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
+                    std::forward<F>(f)(buff);
+                    std::forward<R>(out)(true);
+                    return true;
+                }
+                ipc::yield(k);
+            }
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::multi, trans::broadcast>> {
+
+    using rc_t = std::uint64_t;
+
+    enum : rc_t {
+        ep_mask = 0x00000000ffffffffull,
+        ep_incr = 0x0000000100000000ull
+    };
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<rc_t> rc_ { 0 }; // read-counter
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> wt_;   // write index
+    alignas(cache_line_size) rc_t epoch_ { 0 };             // only one writer
+
+    circ::u2_t cursor() const noexcept {
+        return wt_.load(std::memory_order_acquire);
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & ep_mask;
+            if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch_)) {
+                return false; // has not finished yet
+            }
+            // consider rem_cc to be 0 here
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        std::forward<F>(f)(&(el->data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        epoch_ += ep_incr;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & ep_mask;
+            if (cc & rem_cc) {
+                ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
+                cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
+                if (cc == 0) return false; // no reader
+            }
+            // just compare & exchange
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        std::forward<F>(f)(&(el->data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename R, typename E>
+    bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E* elems) {
+        if (cur == cursor()) return false; // acquire
+        auto* el = elems + circ::index_of(cur++);
+        std::forward<F>(f)(&(el->data_));
+        for (unsigned k = 0;;) {
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            if ((cur_rc & ep_mask) == 0) {
+                std::forward<R>(out)(true);
+                return true;
+            }
+            auto nxt_rc = cur_rc & ~static_cast<rc_t>(wrapper->connected_id());
+            if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
+                std::forward<R>(out)((nxt_rc & ep_mask) == 0);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::multi, relat::multi, trans::broadcast>> {
+
+    using rc_t   = std::uint64_t;
+    using flag_t = std::uint64_t;
+
+    enum : rc_t {
+        rc_mask = 0x00000000ffffffffull,
+        ep_mask = 0x00ffffffffffffffull,
+        ep_incr = 0x0100000000000000ull,
+        ic_mask = 0xff000000ffffffffull,
+        ic_incr = 0x0000000100000000ull
+    };
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<rc_t  > rc_   { 0 }; // read-counter
+        std::atomic<flag_t> f_ct_ { 0 }; // commit flag
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> ct_;   // commit index
+    alignas(cache_line_size) std::atomic<rc_t> epoch_ { 0 };
+
+    circ::u2_t cursor() const noexcept {
+        return ct_.load(std::memory_order_acquire);
+    }
+
+    constexpr static rc_t inc_rc(rc_t rc) noexcept {
+        return (rc & ic_mask) | ((rc + ic_incr) & ~ic_mask);
+    }
+
+    constexpr static rc_t inc_mask(rc_t rc) noexcept {
+        return inc_rc(rc) & ~rc_mask;
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        circ::u2_t cur_ct;
+        rc_t epoch = epoch_.load(std::memory_order_acquire);
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_relaxed);
+            circ::cc_t rem_cc = cur_rc & rc_mask;
+            if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch)) {
+                return false; // has not finished yet
+            }
+            else if (!rem_cc) {
+                auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
+                if ((cur_fl != cur_ct) && cur_fl) {
+                    return false; // full
+                }
+            }
+            // consider rem_cc to be 0 here
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed) &&
+                epoch_.compare_exchange_weak(epoch, epoch, std::memory_order_acq_rel)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        // only one thread/process would touch here at one time
+        ct_.store(cur_ct + 1, std::memory_order_release);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        circ::u2_t cur_ct;
+        rc_t epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & rc_mask;
+            if (cc & rem_cc) {
+                ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
+                cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
+                if (cc == 0) return false; // no reader
+            }
+            // just compare & exchange
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed)) {
+                if (epoch == epoch_.load(std::memory_order_acquire)) {
+                    break;
+                }
+                else if (push(wrapper, std::forward<F>(f), elems)) {
+                    return true;
+                }
+                epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
+            }
+            ipc::yield(k);
+        }
+        // only one thread/process would touch here at one time
+        ct_.store(cur_ct + 1, std::memory_order_release);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename R, typename E, std::size_t N>
+    bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E(& elems)[N]) {
+        auto* el = elems + circ::index_of(cur);
+        auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
+        if (cur_fl != ~static_cast<flag_t>(cur)) {
+            return false; // empty
+        }
+        ++cur;
+        std::forward<F>(f)(&(el->data_));
+        for (unsigned k = 0;;) {
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            if ((cur_rc & rc_mask) == 0) {
+                std::forward<R>(out)(true);
+                el->f_ct_.store(cur + N - 1, std::memory_order_release);
+                return true;
+            }
+            auto nxt_rc = inc_rc(cur_rc) & ~static_cast<rc_t>(wrapper->connected_id());
+            bool last_one = false;
+            if ((last_one = (nxt_rc & rc_mask) == 0)) {
+                el->f_ct_.store(cur + N - 1, std::memory_order_release);
+            }
+            if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
+                std::forward<R>(out)(last_one);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+} // namespace ipc

crazy_functions/test_project/latex/attention/background.tex

@@ -0,0 +1,58 @@
+The goal of reducing sequential computation also forms the foundation of the Extended Neural GPU \citep{extendedngpu}, ByteNet \citep{NalBytenet2017} and ConvS2S \citep{JonasFaceNet2017}, all of which use convolutional neural networks as basic building block, computing hidden representations in parallel for all input and output positions. In these models, the number of operations required to relate signals from two arbitrary input or output positions grows in the distance between positions, linearly for ConvS2S and logarithmically for ByteNet. This makes it more difficult to learn dependencies between distant positions \citep{hochreiter2001gradient}. In the Transformer this is reduced to a constant number of operations, albeit at the cost of reduced effective resolution due to averaging attention-weighted positions, an effect we counteract with Multi-Head Attention as described in section~\ref{sec:attention}. 
+
+Self-attention, sometimes called intra-attention is an attention mechanism relating different positions of a single sequence in order to compute a representation of the sequence. Self-attention has been used successfully in a variety of tasks including reading comprehension, abstractive summarization, textual entailment and learning task-independent sentence representations \citep{cheng2016long, decomposableAttnModel, paulus2017deep, lin2017structured}.
+
+End-to-end memory networks are based on a recurrent attention mechanism instead of sequence-aligned recurrence and have been shown to perform well on simple-language question answering and language modeling tasks \citep{sukhbaatar2015}.
+
+To the best of our knowledge, however, the Transformer is the first transduction model relying entirely on self-attention to compute representations of its input and output without using sequence-aligned RNNs or convolution.
+In the following sections, we will describe the Transformer, motivate self-attention and discuss its advantages over models such as \citep{neural_gpu, NalBytenet2017} and \citep{JonasFaceNet2017}.
+
+
+%\citep{JonasFaceNet2017} report new SOTA on machine translation for English-to-German (EnDe), Enlish-to-French (EnFr) and English-to-Romanian language pairs. 
+
+%For example,! in MT, we must draw information from both input and previous output words to translate an output word accurately. An attention layer \citep{bahdanau2014neural} can connect a very large number of positions at low computation cost, making it an essential ingredient in competitive recurrent models for machine translation.
+
+%A natural question to ask then is, "Could we replace recurrence with attention?". \marginpar{Don't know if it's the most natural question to ask given the previous statements. Also, need to say that the complexity table summarizes these statements} Such a model would be blessed with the computational efficiency of attention and the power of cross-positional communication. In this work, show that pure attention models work remarkably well for MT, achieving new SOTA results on EnDe and EnFr, and can be trained in under $2$ days on xyz architecture. 
+
+%After the seminal models introduced in \citep{sutskever14, bahdanau2014neural, cho2014learning}, recurrent models have become the dominant solution for both sequence modeling and sequence-to-sequence transduction. Many efforts such as \citep{wu2016google,luong2015effective,jozefowicz2016exploring} have pushed the boundaries of machine translation (MT) and language modeling with recurrent endoder-decoder and recurrent language models. Recent effort \citep{shazeer2017outrageously} has successfully combined the power of conditional computation with sequence models to train very large models for MT, pushing SOTA at lower computational cost.
+
+%Recurrent models compute a vector of hidden states $h_t$, for each time step $t$ of computation. $h_t$ is a function of both the input at time $t$ and the previous hidden state $h_t$. This dependence on the previous hidden state precludes processing all timesteps at once, instead requiring long sequences of sequential operations.  In practice, this results in greatly reduced computational efficiency, as on modern computing hardware, a single operation on a large batch is much faster than a large number of operations on small batches.  The problem gets worse at longer sequence lengths. Although sequential computation is not a severe bottleneck at inference time, as autoregressively generating each output requires all previous outputs, the inability to compute scores at all output positions at once hinders us from rapidly training our models over large datasets. Although impressive work such as \citep{Kuchaiev2017Factorization} is able to significantly accelerate the training of LSTMs with factorization tricks, we are still bound by the linear dependence on sequence length.
+
+%If the model could compute hidden states at each time step using only the inputs and outputs,  it would be liberated from the dependence on results from previous time steps during training. This line of thought is the foundation of recent efforts such as the Markovian neural GPU \citep{neural_gpu}, ByteNet \citep{NalBytenet2017} and ConvS2S \citep{JonasFaceNet2017}, all of which use convolutional neural networks as a building block to compute hidden representations simultaneously for all timesteps, resulting in $O(1)$ sequential time complexity. \citep{JonasFaceNet2017} report new SOTA on machine translation for English-to-German (EnDe), Enlish-to-French (EnFr) and English-to-Romanian language pairs. 
+
+%A crucial component for accurate sequence prediction is modeling cross-positional communication. For example, in MT, we must draw information from both input and previous output words to translate an output word accurately. An attention layer \citep{bahdanau2014neural} can connect a very large number of positions at a low computation cost, also $O(1)$ sequential time complexity, making it an essential ingredient in recurrent encoder-decoder architectures for MT. A natural question to ask then is, "Could we replace recurrence with attention?". \marginpar{Don't know if it's the most natural question to ask given the previous statements. Also, need to say that the complexity table summarizes these statements} Such a model would be blessed with the computational efficiency of attention and the power of cross-positional communication. In this work, show that pure attention models work remarkably well for MT, achieving new SOTA results on EnDe and EnFr, and can be trained in under $2$ days on xyz architecture. 
+
+
+
+%Note: Facebook model is no better than RNNs in this regard, since it requires a number of layers proportional to the distance you want to communicate.  Bytenet is more promising, since it requires a logarithmnic number of layers (does bytenet have SOTA results)?   
+
+%Note: An attention  layer can connect a very large number of positions at a low computation cost in O(1) sequential operations.  This is why encoder-decoder attention has been so successful in seq-to-seq models so far.  It is only natural, then, to also use attention to connect the timesteps of the same sequence.
+
+%Note: I wouldn't say that long sequences are not a problem during inference.  It would be great if we could infer with no long sequences.  We could just say later on that, while our training graph is constant-depth, our model still requires sequential operations in the decoder part during inference due to the autoregressive nature of the model.   
+
+%\begin{table}[h!]
+%\caption{Attention models are quite efficient for cross-positional communications when sequence length is smaller than channel depth. $n$ represents the sequence length and $d$ represents the channel depth.}
+%\label{tab:op_complexities}
+%\begin{center}
+%\vspace{-5pt}
+%\scalebox{0.75}{
+
+%\begin{tabular}{l|c|c|c}
+%\hline \hline
+%Layer Type & Receptive & Complexity & Sequential  \\
+%           & Field     &            & Operations  \\
+%\hline
+%Pointwise Feed-Forward & $1$ & $O(n \cdot d^2)$ & $O(1)$ \\
+%\hline
+%Recurrent & $n$ & $O(n \cdot d^2)$ & $O(n)$ \\
+%\hline
+%Convolutional & $r$ & $O(r \cdot n \cdot d^2)$ & $O(1)$ \\
+%\hline
+%Convolutional (separable) & $r$ & $O(r \cdot n \cdot d + n %\cdot d^2)$ & $O(1)$ \\
+%\hline
+%Attention & $r$ & $O(r \cdot n \cdot d)$ & $O(1)$ \\
+%\hline \hline
+%\end{tabular}
+%}
+%\end{center}
+%\end{table}

crazy_functions/test_project/latex/attention/introduction.tex

@@ -0,0 +1,18 @@
+Recurrent neural networks, long short-term memory \citep{hochreiter1997} and gated recurrent \citep{gruEval14} neural networks in particular, have been firmly established as state of the art approaches in sequence modeling and transduction problems such as language modeling and machine translation \citep{sutskever14, bahdanau2014neural, cho2014learning}. Numerous efforts have since continued to push the boundaries of recurrent language models and encoder-decoder architectures \citep{wu2016google,luong2015effective,jozefowicz2016exploring}.
+
+Recurrent models typically factor computation along the symbol positions of the input and output sequences. Aligning the positions to steps in computation time, they generate a sequence of hidden states $h_t$, as a function of the previous hidden state $h_{t-1}$ and the input for position $t$. This inherently sequential nature precludes parallelization within training examples, which becomes critical at longer sequence lengths, as memory constraints limit batching across examples.
+%\marginpar{not sure if the memory constraints are understandable here}
+Recent work has achieved significant improvements in computational efficiency through factorization tricks \citep{Kuchaiev2017Factorization} and conditional computation \citep{shazeer2017outrageously}, while also improving model performance in case of the latter. The fundamental constraint of sequential computation, however, remains.
+
+%\marginpar{@all: there is work on analyzing what attention really does in seq2seq models, couldn't find it right away} 
+
+Attention mechanisms have become an integral part of compelling sequence modeling and transduction models in various tasks, allowing modeling of dependencies without regard to their distance in the input or output sequences \citep{bahdanau2014neural, structuredAttentionNetworks}. In all but a few cases \citep{decomposableAttnModel}, however, such attention mechanisms are used in conjunction with a recurrent network.
+
+%\marginpar{not sure if "cross-positional communication" is understandable without explanation}
+%\marginpar{insert exact training times and stats for the model that reaches sota earliest, maybe even a single GPU model?}
+
+In this work we propose the Transformer, a model architecture eschewing recurrence and instead relying entirely on an attention mechanism to draw global dependencies between input and output. The Transformer allows for significantly more parallelization and can reach a new state of the art in translation quality after being trained for as little as twelve hours on eight P100 GPUs. 
+%\marginpar{you removed the constant number of repetitions part. I wrote it because I wanted to make it clear that the model does not only perform attention once, while it's also not recurrent. I thought that might be important to get across early.}
+
+% Just a standard paragraph with citations, rewrite.
+%After the seminal papers of \citep{sutskever14}, \citep{bahdanau2014neural}, and \citep{cho2014learning}, recurrent models have become the dominant solution for both sequence modeling and sequence-to-sequence transduction. Many efforts such as \citep{wu2016google,luong2015effective,jozefowicz2016exploring} have pushed the boundaries of machine translation and language modeling with recurrent sequence models. Recent effort \citep{shazeer2017outrageously} has combined the power of conditional computation with sequence models to train very large models for machine translation, pushing SOTA at lower computational cost. Recurrent models compute a vector of hidden states $h_t$, for each time step $t$ of computation. $h_t$ is a function of both the input at time $t$ and the previous hidden state $h_t$. This dependence on the previous hidden state encumbers recurrnet models to process multiple inputs at once, and their time complexity is a linear function of the length of the input and output, both during training and inference. [What I want to say here is that although this is fine during decoding, at training time, we are given both input and output and this linear nature does not allow the RNN to process all inputs and outputs simultaneously and haven't been used on datasets that are the of the scale of the web. What's the largest dataset we have ? . Talk about Nividia and possibly other's effors to speed up things, and possibly other efforts that alleviate this, but are still limited by it's comptuational nature]. Rest of the intro: What if you could construct the state based on the actual inputs and outputs, then you could construct them all at once. This has been the foundation of many promising recent efforts, bytenet,facenet (Also talk about quasi rnn here). Now we talk about attention!! Along with cell architectures such as long short-term meory (LSTM) \citep{hochreiter1997}, and gated recurrent units (GRUs) \citep{cho2014learning}, attention has emerged as an essential ingredient in successful sequence models, in particular for machine translation. In recent years, many, if not all, state-of-the-art (SOTA) results in machine translation have been achieved with attention-based sequence models \citep{wu2016google,luong2015effective,jozefowicz2016exploring}. Talk about the neon work on how it played with attention to do self attention! Then talk about what we do.

crazy_functions/test_project/latex/attention/model_architecture.tex

@@ -0,0 +1,155 @@
+
+\begin{figure}
+  \centering
+  \includegraphics[scale=0.6]{Figures/ModalNet-21}
+  \caption{The Transformer - model architecture.}
+  \label{fig:model-arch}
+\end{figure}
+
+% Although the primary workhorse of our model is attention, 
+%Our model maintains the encoder-decoder structure that is common to many so-called sequence-to-sequence models \citep{bahdanau2014neural,sutskever14}.  As in all such architectures, the encoder computes a representation of the input sequence, and the decoder consumes these representations along with the output tokens to autoregressively produce the output sequence.  Where, traditionally, the encoder and decoder contain stacks of recurrent or convolutional layers, our encoder and decoder stacks are composed of attention layers and position-wise feed-forward layers (Figure~\ref{fig:model-arch}).  The following sections describe the gross architecture and these particular components in detail.
+
+Most competitive neural sequence transduction models have an encoder-decoder structure \citep{cho2014learning,bahdanau2014neural,sutskever14}. Here, the encoder maps an input sequence of symbol representations $(x_1, ..., x_n)$ to a sequence of continuous representations $\mathbf{z} = (z_1, ..., z_n)$. Given $\mathbf{z}$, the decoder then generates an output sequence $(y_1,...,y_m)$ of symbols one element at a time. At each step the model is auto-regressive \citep{graves2013generating}, consuming the previously generated symbols as additional input when generating the next.
+
+The Transformer follows this overall architecture using stacked self-attention and point-wise, fully connected layers for both the encoder and decoder, shown in the left and right halves of Figure~\ref{fig:model-arch}, respectively.
+
+\subsection{Encoder and Decoder Stacks}
+
+\paragraph{Encoder:}The encoder is composed of a stack of $N=6$ identical layers. Each layer has two sub-layers. The first is a multi-head self-attention mechanism, and the second is a simple, position-wise fully connected feed-forward network.   We employ a residual connection \citep{he2016deep} around each of the two sub-layers, followed by layer normalization \cite{layernorm2016}.  That is, the output of each sub-layer is $\mathrm{LayerNorm}(x + \mathrm{Sublayer}(x))$, where $\mathrm{Sublayer}(x)$ is the function implemented by the sub-layer itself.  To facilitate these residual connections, all sub-layers in the model, as well as the embedding layers, produce outputs of dimension $\dmodel=512$.
+
+\paragraph{Decoder:}The decoder is also composed of a stack of $N=6$ identical layers.  In addition to the two sub-layers in each encoder layer, the decoder inserts a third sub-layer, which performs multi-head attention over the output of the encoder stack.  Similar to the encoder, we employ residual connections around each of the sub-layers, followed by layer normalization.  We also modify the self-attention sub-layer in the decoder stack to prevent positions from attending to subsequent positions.  This masking, combined with fact that the output embeddings are offset by one position, ensures that the predictions for position $i$ can depend only on the known outputs at positions less than $i$.
+
+% In our model (Figure~\ref{fig:model-arch}), the encoder and decoder are composed of stacks of alternating self-attention layers (for cross-positional communication) and position-wise feed-forward layers (for in-place computation).  In addition, the decoder stack contains encoder-decoder attention layers.  Since attention is agnostic to the distances between words, our model requires a "positional encoding" to be added to the encoder and decoder input. The following sections describe all of these components in detail.
+
+\subsection{Attention} \label{sec:attention}
+An attention function can be described as mapping a query and a set of key-value pairs to an output, where the query, keys, values, and output are all vectors.  The output is computed as a weighted sum of the values, where the weight assigned to each value is computed by a compatibility function of the query with the corresponding key.
+
+\subsubsection{Scaled Dot-Product Attention} \label{sec:scaled-dot-prod}
+
+% \begin{figure}
+%   \centering
+%   \includegraphics[scale=0.6]{Figures/ModalNet-19}
+%   \caption{Scaled Dot-Product Attention.}
+%   \label{fig:multi-head-att}
+% \end{figure}
+
+We call our particular attention "Scaled Dot-Product Attention" (Figure~\ref{fig:multi-head-att}).   The input consists of queries and keys of dimension $d_k$, and values of dimension $d_v$.  We compute the dot products of the query with all keys, divide each by $\sqrt{d_k}$, and apply a softmax function to obtain the weights on the values.
+
+In practice, we compute the attention function on a set of queries simultaneously, packed together into a matrix $Q$.   The keys and values are also packed together into matrices $K$ and $V$.  We compute the matrix of outputs as:
+
+\begin{equation}
+   \mathrm{Attention}(Q, K, V) = \mathrm{softmax}(\frac{QK^T}{\sqrt{d_k}})V
+\end{equation}
+
+The two most commonly used attention functions are additive attention \citep{bahdanau2014neural}, and dot-product (multiplicative) attention.  Dot-product attention is identical to our algorithm, except for the scaling factor of $\frac{1}{\sqrt{d_k}}$. Additive attention computes the compatibility function using a feed-forward network with a single hidden layer.  While the two are similar in theoretical complexity, dot-product attention is much faster and more space-efficient in practice, since it can be implemented using highly optimized matrix multiplication code. 
+
+%We scale the dot products by $1/\sqrt{d_k}$ to limit the magnitude of the dot products, which works well in practice. Otherwise, we found applying the softmax to often result in weights very close to 0 or 1, and hence minuscule gradients.
+
+% Already described in the subsequent section
+%When used as part of decoder self-attention, an optional mask function is applied just before the softmax to prevent positions from attending to subsequent positions.   This mask simply sets the logits corresponding to all illegal connections (those outside of the lower triangle) to $-\infty$.
+
+%\paragraph{Comparison to Additive Attention: } We choose dot product attention over additive attention \citep{bahdanau2014neural} since it can be computed using highly optimized matrix multiplication code.  This optimization is particularly important to us, as we employ many attention layers in our model.
+
+While for small values of $d_k$ the two mechanisms perform similarly, additive attention outperforms dot product attention without scaling for larger values of $d_k$ \citep{DBLP:journals/corr/BritzGLL17}. We suspect that for large values of $d_k$, the dot products grow large in magnitude, pushing the softmax function into regions where it has extremely small gradients  \footnote{To illustrate why the dot products get large, assume that the components of $q$ and $k$ are independent random variables with mean $0$ and variance $1$.  Then their dot product, $q \cdot k = \sum_{i=1}^{d_k} q_ik_i$, has mean $0$ and variance $d_k$.}. To counteract this effect, we scale the dot products by $\frac{1}{\sqrt{d_k}}$.
+
+
+%We suspect this to be caused by the dot products growing too large in magnitude to result in useful gradients after applying the softmax function.  To counteract this, we scale the dot product by $1/\sqrt{d_k}$.
+
+
+\subsubsection{Multi-Head Attention} \label{sec:multihead}
+
+\begin{figure}
+\begin{minipage}[t]{0.5\textwidth}
+  \centering
+  Scaled Dot-Product Attention \\
+  \vspace{0.5cm}
+  \includegraphics[scale=0.6]{Figures/ModalNet-19}
+\end{minipage}
+\begin{minipage}[t]{0.5\textwidth}
+  \centering 
+  Multi-Head Attention \\
+  \vspace{0.1cm}
+  \includegraphics[scale=0.6]{Figures/ModalNet-20}  
+\end{minipage}
+
+
+  % \centering
+
+  \caption{(left) Scaled Dot-Product Attention. (right) Multi-Head Attention consists of several attention layers running in parallel.}
+  \label{fig:multi-head-att}
+\end{figure}
+
+Instead of performing a single attention function with $\dmodel$-dimensional keys, values and queries, we found it beneficial to linearly project the queries, keys and values $h$ times with different, learned linear projections to $d_k$, $d_k$ and $d_v$ dimensions, respectively.
+On each of these projected versions of queries, keys and values we then perform the attention function in parallel, yielding $d_v$-dimensional output values. These are concatenated and once again projected, resulting in the final values, as depicted in Figure~\ref{fig:multi-head-att}.
+
+Multi-head attention allows the model to jointly attend to information from different representation subspaces at different positions. With a single attention head, averaging inhibits this.
+
+\begin{align*}
+    \mathrm{MultiHead}(Q, K, V) &= \mathrm{Concat}(\mathrm{head_1}, ..., \mathrm{head_h})W^O\\
+%    \mathrm{where} \mathrm{head_i} &= \mathrm{Attention}(QW_Q_i^{\dmodel \times d_q}, KW_K_i^{\dmodel \times d_k}, VW^V_i^{\dmodel \times d_v})\\
+    \text{where}~\mathrm{head_i} &= \mathrm{Attention}(QW^Q_i, KW^K_i, VW^V_i)\\
+\end{align*}
+
+Where the projections are parameter matrices $W^Q_i \in \mathbb{R}^{\dmodel \times d_k}$, $W^K_i \in \mathbb{R}^{\dmodel \times d_k}$, $W^V_i \in \mathbb{R}^{\dmodel \times d_v}$ and $W^O \in \mathbb{R}^{hd_v \times \dmodel}$.
+
+
+%find it better (and no more expensive) to have multiple parallel attention layers (each over the full set of positions) with proportionally lower-dimensional keys, values and queries.  We call this "Multi-Head Attention" (Figure~\ref{fig:multi-head-att}).  The keys, values, and queries for each of these parallel attention layers are computed by learned linear transformations of the inputs to the multi-head attention.  We use different linear transformations across different parallel attention layers.  The output of the parallel attention layers are concatenated, and then passed through a final learned linear transformation. 
+
+In this work we employ $h=8$ parallel attention layers, or heads. For each of these we use $d_k=d_v=\dmodel/h=64$.
+Due to the reduced dimension of each head, the total computational cost is similar to that of single-head attention with full dimensionality.
+
+\subsubsection{Applications of Attention in our Model}
+
+The Transformer uses multi-head attention in three different ways: 
+\begin{itemize}
+ \item In "encoder-decoder attention" layers, the queries come from the previous decoder layer, and the memory keys and values come from the output of the encoder.   This allows every position in the decoder to attend over all positions in the input sequence.  This mimics the typical encoder-decoder attention mechanisms in sequence-to-sequence models such as \citep{wu2016google, bahdanau2014neural,JonasFaceNet2017}.
+
+ \item The encoder contains self-attention layers.  In a self-attention layer all of the keys, values and queries come from the same place, in this case, the output of the previous layer in the encoder.   Each position in the encoder can attend to all positions in the previous layer of the encoder.
+
+ \item Similarly, self-attention layers in the decoder allow each position in the decoder to attend to all positions in the decoder up to and including that position.  We need to prevent leftward information flow in the decoder to preserve the auto-regressive property.  We implement this inside of scaled dot-product attention by masking out (setting to $-\infty$) all values in the input of the softmax which correspond to illegal connections.  See Figure~\ref{fig:multi-head-att}.
+
+\end{itemize}
+
+\subsection{Position-wise Feed-Forward Networks}\label{sec:ffn}
+
+In addition to attention sub-layers, each of the layers in our encoder and decoder contains a fully connected feed-forward network, which is applied to each position separately and identically.  This consists of two linear transformations with a ReLU activation in between.
+
+\begin{equation}
+   \mathrm{FFN}(x)=\max(0, xW_1 + b_1) W_2 + b_2
+\end{equation}
+
+While the linear transformations are the same across different positions, they use different parameters from layer to layer. Another way of describing this is as two convolutions with kernel size 1.  The dimensionality of input and output is $\dmodel=512$, and the inner-layer has dimensionality $d_{ff}=2048$.
+
+
+
+%In the appendix, we describe how the position-wise feed-forward network can also be seen as a form of attention.
+
+%from Jakob: The number of operations required for the model to relate signals from two arbitrary input or output positions grows in the distance between positions in input or output, linearly for ConvS2S and logarithmically for ByteNet, making it harder to learn dependencies between these positions \citep{hochreiter2001gradient}. In the transformer this is reduced to a constant number of operations, albeit at the cost of effective resolution caused by averaging attention-weighted positions, an effect we aim to counteract with multi-headed attention.
+
+
+%Figure~\ref{fig:simple-att} presents a simple attention function, $A$, with a single head, that forms the basis of our multi-head attention. $A$ takes a query key vector $\kq$, matrices of memory keys $\km$ and memory values $\vm$ ,and produces a query value vector $\vq$ as 
+%\begin{equation*} \label{eq:attention}
+%    A(\kq, \km, \vm) = {\vm}^T (Softmax(\km \kq).
+%\end{equation*}
+%We linearly transform $\kq,\,\km$, and $\vm$ with learned matrices ${\Wkq \text{,} \, \Wkm}$, and ${\Wvm}$ before calling the attention function, and transform the output query with $\Wvq$ before handing it to the feed forward layer. Each attention layer has it's own set of transformation matrices, which are shared across all query positions. $A$ is applied in parallel for each query position, and is implemented very efficiently as a batch of matrix multiplies. The self-attention and encoder-decoder attention layers use $A$, but with different arguments. For example, in encdoder self-attention, queries in encoder layer $i$ attention to memories in encoder layer $i-1$. To ensure that decoder self-attention layers do not look at future words, we add $- \inf$ to the softmax logits in positions $j+1$ to query length for query position $l$.  
+
+%In simple attention, the query value is a weighted combination of the memory values where the attention weights sum to one. Although this function performs well in practice, the constraint on attention weights can restrict the amount of information that flows from memories to queries because the query cannot focus on multiple memory positions at once, which might be desirable when translating long sequences. \marginpar{@usz, could you think of an example of this ?} We remedy this by maintaining multiple attention heads at each query position that attend to all memory positions in parallel, with a different set of parameters  per attention head $h$. 
+%\marginpar{}
+
+\subsection{Embeddings and Softmax}
+Similarly to other sequence transduction models, we use learned embeddings to convert the input tokens and output tokens to vectors of dimension $\dmodel$.  We also use the usual learned linear transformation and softmax function to convert the decoder output to predicted next-token probabilities.  In our model, we share the same weight matrix between the two embedding layers and the pre-softmax linear transformation, similar to \citep{press2016using}.   In the embedding layers, we multiply those weights by $\sqrt{\dmodel}$.
+
+
+\subsection{Positional Encoding}
+Since our model contains no recurrence and no convolution, in order for the model to make use of the order of the sequence, we must inject some information about the relative or absolute position of the tokens in the sequence.  To this end, we add "positional encodings" to the input embeddings at the bottoms of the encoder and decoder stacks.  The positional encodings have the same dimension $\dmodel$ as the embeddings, so that the two can be summed.   There are many choices of positional encodings, learned and fixed \citep{JonasFaceNet2017}.
+
+In this work, we use sine and cosine functions of different frequencies:
+
+\begin{align*}
+    PE_{(pos,2i)} = sin(pos / 10000^{2i/\dmodel}) \\
+    PE_{(pos,2i+1)} = cos(pos / 10000^{2i/\dmodel})
+\end{align*}
+
+where $pos$ is the position and $i$ is the dimension.  That is, each dimension of the positional encoding corresponds to a sinusoid.  The wavelengths form a geometric progression from $2\pi$ to $10000 \cdot 2\pi$.  We chose this function because we hypothesized it would allow the model to easily learn to attend by relative positions, since for any fixed offset $k$, $PE_{pos+k}$ can be represented as a linear function of $PE_{pos}$.
+
+We also experimented with using learned positional embeddings \citep{JonasFaceNet2017} instead, and found that the two versions produced nearly identical results (see Table~\ref{tab:variations} row (E)).  We chose the sinusoidal version because it may allow the model to extrapolate to sequence lengths longer than the ones encountered during training.

crazy_functions/test_project/latex/attention/parameter_attention.tex

@@ -0,0 +1,45 @@
+\pagebreak
+\section*{Two Feed-Forward Layers = Attention over Parameters}\label{sec:parameter_attention}
+
+In addition to attention layers, our model contains position-wise feed-forward networks (Section \ref{sec:ffn}), which consist of two linear transformations with a ReLU activation in between.  In fact, these networks too can be seen as a form of attention.  Compare the formula for such a network with the formula for a simple dot-product attention layer (biases and scaling factors omitted):
+
+\begin{align*}
+    FFN(x, W_1, W_2) = ReLU(xW_1)W_2 \\
+    A(q, K, V) = Softmax(qK^T)V
+\end{align*}
+
+Based on the similarity of these formulae, the two-layer feed-forward network can be seen as a kind of attention, where the keys and values are the rows of the trainable parameter matrices $W_1$ and $W_2$, and where we use ReLU instead of Softmax in the compatibility function.
+
+%the compatablity function is $compat(q, k_i) = ReLU(q \cdot k_i)$ instead of $Softmax(qK_T)_i$.
+
+Given this similarity, we experimented with replacing the position-wise feed-forward networks with attention layers similar to the ones we use everywhere else our model. The multi-head-attention-over-parameters sublayer is identical to the multi-head attention described in \ref{sec:multihead}, except that the "keys" and "values" inputs to each attention head are trainable model parameters, as opposed to being linear projections of a previous layer.  These parameters are scaled up by a factor of $\sqrt{d_{model}}$ in order to be more similar to activations.
+
+In our first experiment, we replaced each position-wise feed-forward network with a multi-head-attention-over-parameters sublayer with $h_p=8$ heads, key-dimensionality $d_{pk}=64$, and value-dimensionality $d_{pv}=64$, using $n_p=1536$ key-value pairs for each attention head.  The sublayer has a total of $2097152$ parameters, including the parameters in the query projection and the output projection.  This matches the number of parameters in the position-wise feed-forward network that we replaced.  While the theoretical amount of computation is also the same, in practice, the attention version caused the step times to be about 30\% longer.
+
+In our second experiment, we used $h_p=8$ heads, and $n_p=512$ key-value pairs for each attention head, again matching the total number of parameters in the base model.
+
+Results for the first experiment were slightly worse than for the base model, and results for the second experiment were slightly better, see Table~\ref{tab:parameter_attention}.
+
+\begin{table}[h]
+\caption{Replacing the position-wise feed-forward networks with multihead-attention-over-parameters produces similar results to the base model.  All metrics are on the English-to-German translation development set, newstest2013.}
+\label{tab:parameter_attention}
+\begin{center}
+\vspace{-2mm}
+%\scalebox{1.0}{
+\begin{tabular}{c|cccccc|cccc}
+\hline\rule{0pt}{2.0ex}
+ & \multirow{2}{*}{$\dmodel$} & \multirow{2}{*}{$\dff$} &
+\multirow{2}{*}{$h_p$} & \multirow{2}{*}{$d_{pk}$} & \multirow{2}{*}{$d_{pv}$} &
+ \multirow{2}{*}{$n_p$} &
+ PPL & BLEU & params & training\\
+ & & & & & &  & (dev) & (dev) & $\times10^6$ & time \\
+\hline\rule{0pt}{2.0ex}
+base & 512 & 2048 & & & & & 4.92 & 25.8 & 65 & 12 hours\\
+\hline\rule{0pt}{2.0ex}
+AOP$_1$ & 512 & & 8 & 64 & 64 & 1536 & 4.92& 25.5  & 65 & 16 hours\\
+AOP$_2$ & 512 & & 16 & 64 & 64 & 512 & \textbf{4.86} & \textbf{25.9}  & 65 & 16 hours \\
+\hline
+\end{tabular}
+%}
+\end{center}
+\end{table}

crazy_functions/test_project/latex/attention/来源

@@ -0,0 +1,8 @@
+chatgpt的老祖宗《Attention is all you need》
+
+Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin
+
+真实的摘要如下
+The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.
+
+https://arxiv.org/abs/1706.03762

crazy_functions/test_project/python/dqn/__init__.py

@@ -0,0 +1,2 @@
+from stable_baselines3.dqn.dqn import DQN
+from stable_baselines3.dqn.policies import CnnPolicy, MlpPolicy

crazy_functions/test_project/python/dqn/dqn.py

@@ -0,0 +1,245 @@
+from typing import Any, Dict, List, Optional, Tuple, Type, Union
+
+import gym
+import numpy as np
+import torch as th
+from torch.nn import functional as F
+
+from stable_baselines3.common import logger
+from stable_baselines3.common.off_policy_algorithm import OffPolicyAlgorithm
+from stable_baselines3.common.preprocessing import maybe_transpose
+from stable_baselines3.common.type_aliases import GymEnv, MaybeCallback, Schedule
+from stable_baselines3.common.utils import get_linear_fn, is_vectorized_observation, polyak_update
+from stable_baselines3.dqn.policies import DQNPolicy
+
+
+class DQN(OffPolicyAlgorithm):
+    """
+    Deep Q-Network (DQN)
+
+    Paper: https://arxiv.org/abs/1312.5602, https://www.nature.com/articles/nature14236
+    Default hyperparameters are taken from the nature paper,
+    except for the optimizer and learning rate that were taken from Stable Baselines defaults.
+
+    :param policy: The policy model to use (MlpPolicy, CnnPolicy, ...)
+    :param env: The environment to learn from (if registered in Gym, can be str)
+    :param learning_rate: The learning rate, it can be a function
+        of the current progress remaining (from 1 to 0)
+    :param buffer_size: size of the replay buffer
+    :param learning_starts: how many steps of the model to collect transitions for before learning starts
+    :param batch_size: Minibatch size for each gradient update
+    :param tau: the soft update coefficient ("Polyak update", between 0 and 1) default 1 for hard update
+    :param gamma: the discount factor
+    :param train_freq: Update the model every ``train_freq`` steps. Alternatively pass a tuple of frequency and unit
+        like ``(5, "step")`` or ``(2, "episode")``.
+    :param gradient_steps: How many gradient steps to do after each rollout (see ``train_freq``)
+        Set to ``-1`` means to do as many gradient steps as steps done in the environment
+        during the rollout.
+    :param optimize_memory_usage: Enable a memory efficient variant of the replay buffer
+        at a cost of more complexity.
+        See https://github.com/DLR-RM/stable-baselines3/issues/37#issuecomment-637501195
+    :param target_update_interval: update the target network every ``target_update_interval``
+        environment steps.
+    :param exploration_fraction: fraction of entire training period over which the exploration rate is reduced
+    :param exploration_initial_eps: initial value of random action probability
+    :param exploration_final_eps: final value of random action probability
+    :param max_grad_norm: The maximum value for the gradient clipping
+    :param tensorboard_log: the log location for tensorboard (if None, no logging)
+    :param create_eval_env: Whether to create a second environment that will be
+        used for evaluating the agent periodically. (Only available when passing string for the environment)
+    :param policy_kwargs: additional arguments to be passed to the policy on creation
+    :param verbose: the verbosity level: 0 no output, 1 info, 2 debug
+    :param seed: Seed for the pseudo random generators
+    :param device: Device (cpu, cuda, ...) on which the code should be run.
+        Setting it to auto, the code will be run on the GPU if possible.
+    :param _init_setup_model: Whether or not to build the network at the creation of the instance
+    """
+
+    def __init__(
+        self,
+        policy: Union[str, Type[DQNPolicy]],
+        env: Union[GymEnv, str],
+        learning_rate: Union[float, Schedule] = 1e-4,
+        buffer_size: int = 1000000,
+        learning_starts: int = 50000,
+        batch_size: Optional[int] = 32,
+        tau: float = 1.0,
+        gamma: float = 0.99,
+        train_freq: Union[int, Tuple[int, str]] = 4,
+        gradient_steps: int = 1,
+        optimize_memory_usage: bool = False,
+        target_update_interval: int = 10000,
+        exploration_fraction: float = 0.1,
+        exploration_initial_eps: float = 1.0,
+        exploration_final_eps: float = 0.05,
+        max_grad_norm: float = 10,
+        tensorboard_log: Optional[str] = None,
+        create_eval_env: bool = False,
+        policy_kwargs: Optional[Dict[str, Any]] = None,
+        verbose: int = 0,
+        seed: Optional[int] = None,
+        device: Union[th.device, str] = "auto",
+        _init_setup_model: bool = True,
+    ):
+
+        super(DQN, self).__init__(
+            policy,
+            env,
+            DQNPolicy,
+            learning_rate,
+            buffer_size,
+            learning_starts,
+            batch_size,
+            tau,
+            gamma,
+            train_freq,
+            gradient_steps,
+            action_noise=None,  # No action noise
+            policy_kwargs=policy_kwargs,
+            tensorboard_log=tensorboard_log,
+            verbose=verbose,
+            device=device,
+            create_eval_env=create_eval_env,
+            seed=seed,
+            sde_support=False,
+            optimize_memory_usage=optimize_memory_usage,
+            supported_action_spaces=(gym.spaces.Discrete,),
+        )
+
+        self.exploration_initial_eps = exploration_initial_eps
+        self.exploration_final_eps = exploration_final_eps
+        self.exploration_fraction = exploration_fraction
+        self.target_update_interval = target_update_interval
+        self.max_grad_norm = max_grad_norm
+        # "epsilon" for the epsilon-greedy exploration
+        self.exploration_rate = 0.0
+        # Linear schedule will be defined in `_setup_model()`
+        self.exploration_schedule = None
+        self.q_net, self.q_net_target = None, None
+
+        if _init_setup_model:
+            self._setup_model()
+
+    def _setup_model(self) -> None:
+        super(DQN, self)._setup_model()
+        self._create_aliases()
+        self.exploration_schedule = get_linear_fn(
+            self.exploration_initial_eps, self.exploration_final_eps, self.exploration_fraction
+        )
+
+    def _create_aliases(self) -> None:
+        self.q_net = self.policy.q_net
+        self.q_net_target = self.policy.q_net_target
+
+    def _on_step(self) -> None:
+        """
+        Update the exploration rate and target network if needed.
+        This method is called in ``collect_rollouts()`` after each step in the environment.
+        """
+        if self.num_timesteps % self.target_update_interval == 0:
+            polyak_update(self.q_net.parameters(), self.q_net_target.parameters(), self.tau)
+
+        self.exploration_rate = self.exploration_schedule(self._current_progress_remaining)
+        logger.record("rollout/exploration rate", self.exploration_rate)
+
+    def train(self, gradient_steps: int, batch_size: int = 100) -> None:
+        # Update learning rate according to schedule
+        self._update_learning_rate(self.policy.optimizer)
+
+        losses = []
+        for _ in range(gradient_steps):
+            # Sample replay buffer
+            replay_data = self.replay_buffer.sample(batch_size, env=self._vec_normalize_env)
+
+            with th.no_grad():
+                # Compute the next Q-values using the target network
+                next_q_values = self.q_net_target(replay_data.next_observations)
+                # Follow greedy policy: use the one with the highest value
+                next_q_values, _ = next_q_values.max(dim=1)
+                # Avoid potential broadcast issue
+                next_q_values = next_q_values.reshape(-1, 1)
+                # 1-step TD target
+                target_q_values = replay_data.rewards + (1 - replay_data.dones) * self.gamma * next_q_values
+
+            # Get current Q-values estimates
+            current_q_values = self.q_net(replay_data.observations)
+
+            # Retrieve the q-values for the actions from the replay buffer
+            current_q_values = th.gather(current_q_values, dim=1, index=replay_data.actions.long())
+
+            # Compute Huber loss (less sensitive to outliers)
+            loss = F.smooth_l1_loss(current_q_values, target_q_values)
+            losses.append(loss.item())
+
+            # Optimize the policy
+            self.policy.optimizer.zero_grad()
+            loss.backward()
+            # Clip gradient norm
+            th.nn.utils.clip_grad_norm_(self.policy.parameters(), self.max_grad_norm)
+            self.policy.optimizer.step()
+
+        # Increase update counter
+        self._n_updates += gradient_steps
+
+        logger.record("train/n_updates", self._n_updates, exclude="tensorboard")
+        logger.record("train/loss", np.mean(losses))
+
+    def predict(
+        self,
+        observation: np.ndarray,
+        state: Optional[np.ndarray] = None,
+        mask: Optional[np.ndarray] = None,
+        deterministic: bool = False,
+    ) -> Tuple[np.ndarray, Optional[np.ndarray]]:
+        """
+        Overrides the base_class predict function to include epsilon-greedy exploration.
+
+        :param observation: the input observation
+        :param state: The last states (can be None, used in recurrent policies)
+        :param mask: The last masks (can be None, used in recurrent policies)
+        :param deterministic: Whether or not to return deterministic actions.
+        :return: the model's action and the next state
+            (used in recurrent policies)
+        """
+        if not deterministic and np.random.rand() < self.exploration_rate:
+            if is_vectorized_observation(maybe_transpose(observation, self.observation_space), self.observation_space):
+                n_batch = observation.shape[0]
+                action = np.array([self.action_space.sample() for _ in range(n_batch)])
+            else:
+                action = np.array(self.action_space.sample())
+        else:
+            action, state = self.policy.predict(observation, state, mask, deterministic)
+        return action, state
+
+    def learn(
+        self,
+        total_timesteps: int,
+        callback: MaybeCallback = None,
+        log_interval: int = 4,
+        eval_env: Optional[GymEnv] = None,
+        eval_freq: int = -1,
+        n_eval_episodes: int = 5,
+        tb_log_name: str = "DQN",
+        eval_log_path: Optional[str] = None,
+        reset_num_timesteps: bool = True,
+    ) -> OffPolicyAlgorithm:
+
+        return super(DQN, self).learn(
+            total_timesteps=total_timesteps,
+            callback=callback,
+            log_interval=log_interval,
+            eval_env=eval_env,
+            eval_freq=eval_freq,
+            n_eval_episodes=n_eval_episodes,
+            tb_log_name=tb_log_name,
+            eval_log_path=eval_log_path,
+            reset_num_timesteps=reset_num_timesteps,
+        )
+
+    def _excluded_save_params(self) -> List[str]:
+        return super(DQN, self)._excluded_save_params() + ["q_net", "q_net_target"]
+
+    def _get_torch_save_params(self) -> Tuple[List[str], List[str]]:
+        state_dicts = ["policy", "policy.optimizer"]
+
+        return state_dicts, []

crazy_functions/test_project/python/dqn/policies.py

@@ -0,0 +1,237 @@
+from typing import Any, Dict, List, Optional, Type
+
+import gym
+import torch as th
+from torch import nn
+
+from stable_baselines3.common.policies import BasePolicy, register_policy
+from stable_baselines3.common.torch_layers import BaseFeaturesExtractor, FlattenExtractor, NatureCNN, create_mlp
+from stable_baselines3.common.type_aliases import Schedule
+
+
+class QNetwork(BasePolicy):
+    """
+    Action-Value (Q-Value) network for DQN
+
+    :param observation_space: Observation space
+    :param action_space: Action space
+    :param net_arch: The specification of the policy and value networks.
+    :param activation_fn: Activation function
+    :param normalize_images: Whether to normalize images or not,
+         dividing by 255.0 (True by default)
+    """
+
+    def __init__(
+        self,
+        observation_space: gym.spaces.Space,
+        action_space: gym.spaces.Space,
+        features_extractor: nn.Module,
+        features_dim: int,
+        net_arch: Optional[List[int]] = None,
+        activation_fn: Type[nn.Module] = nn.ReLU,
+        normalize_images: bool = True,
+    ):
+        super(QNetwork, self).__init__(
+            observation_space,
+            action_space,
+            features_extractor=features_extractor,
+            normalize_images=normalize_images,
+        )
+
+        if net_arch is None:
+            net_arch = [64, 64]
+
+        self.net_arch = net_arch
+        self.activation_fn = activation_fn
+        self.features_extractor = features_extractor
+        self.features_dim = features_dim
+        self.normalize_images = normalize_images
+        action_dim = self.action_space.n  # number of actions
+        q_net = create_mlp(self.features_dim, action_dim, self.net_arch, self.activation_fn)
+        self.q_net = nn.Sequential(*q_net)
+
+    def forward(self, obs: th.Tensor) -> th.Tensor:
+        """
+        Predict the q-values.
+
+        :param obs: Observation
+        :return: The estimated Q-Value for each action.
+        """
+        return self.q_net(self.extract_features(obs))
+
+    def _predict(self, observation: th.Tensor, deterministic: bool = True) -> th.Tensor:
+        q_values = self.forward(observation)
+        # Greedy action
+        action = q_values.argmax(dim=1).reshape(-1)
+        return action
+
+    def _get_constructor_parameters(self) -> Dict[str, Any]:
+        data = super()._get_constructor_parameters()
+
+        data.update(
+            dict(
+                net_arch=self.net_arch,
+                features_dim=self.features_dim,
+                activation_fn=self.activation_fn,
+                features_extractor=self.features_extractor,
+            )
+        )
+        return data
+
+
+class DQNPolicy(BasePolicy):
+    """
+    Policy class with Q-Value Net and target net for DQN
+
+    :param observation_space: Observation space
+    :param action_space: Action space
+    :param lr_schedule: Learning rate schedule (could be constant)
+    :param net_arch: The specification of the policy and value networks.
+    :param activation_fn: Activation function
+    :param features_extractor_class: Features extractor to use.
+    :param features_extractor_kwargs: Keyword arguments
+        to pass to the features extractor.
+    :param normalize_images: Whether to normalize images or not,
+         dividing by 255.0 (True by default)
+    :param optimizer_class: The optimizer to use,
+        ``th.optim.Adam`` by default
+    :param optimizer_kwargs: Additional keyword arguments,
+        excluding the learning rate, to pass to the optimizer
+    """
+
+    def __init__(
+        self,
+        observation_space: gym.spaces.Space,
+        action_space: gym.spaces.Space,
+        lr_schedule: Schedule,
+        net_arch: Optional[List[int]] = None,
+        activation_fn: Type[nn.Module] = nn.ReLU,
+        features_extractor_class: Type[BaseFeaturesExtractor] = FlattenExtractor,
+        features_extractor_kwargs: Optional[Dict[str, Any]] = None,
+        normalize_images: bool = True,
+        optimizer_class: Type[th.optim.Optimizer] = th.optim.Adam,
+        optimizer_kwargs: Optional[Dict[str, Any]] = None,
+    ):
+        super(DQNPolicy, self).__init__(
+            observation_space,
+            action_space,
+            features_extractor_class,
+            features_extractor_kwargs,
+            optimizer_class=optimizer_class,
+            optimizer_kwargs=optimizer_kwargs,
+        )
+
+        if net_arch is None:
+            if features_extractor_class == FlattenExtractor:
+                net_arch = [64, 64]
+            else:
+                net_arch = []
+
+        self.net_arch = net_arch
+        self.activation_fn = activation_fn
+        self.normalize_images = normalize_images
+
+        self.net_args = {
+            "observation_space": self.observation_space,
+            "action_space": self.action_space,
+            "net_arch": self.net_arch,
+            "activation_fn": self.activation_fn,
+            "normalize_images": normalize_images,
+        }
+
+        self.q_net, self.q_net_target = None, None
+        self._build(lr_schedule)
+
+    def _build(self, lr_schedule: Schedule) -> None:
+        """
+        Create the network and the optimizer.
+
+        :param lr_schedule: Learning rate schedule
+            lr_schedule(1) is the initial learning rate
+        """
+
+        self.q_net = self.make_q_net()
+        self.q_net_target = self.make_q_net()
+        self.q_net_target.load_state_dict(self.q_net.state_dict())
+
+        # Setup optimizer with initial learning rate
+        self.optimizer = self.optimizer_class(self.parameters(), lr=lr_schedule(1), **self.optimizer_kwargs)
+
+    def make_q_net(self) -> QNetwork:
+        # Make sure we always have separate networks for features extractors etc
+        net_args = self._update_features_extractor(self.net_args, features_extractor=None)
+        return QNetwork(**net_args).to(self.device)
+
+    def forward(self, obs: th.Tensor, deterministic: bool = True) -> th.Tensor:
+        return self._predict(obs, deterministic=deterministic)
+
+    def _predict(self, obs: th.Tensor, deterministic: bool = True) -> th.Tensor:
+        return self.q_net._predict(obs, deterministic=deterministic)
+
+    def _get_constructor_parameters(self) -> Dict[str, Any]:
+        data = super()._get_constructor_parameters()
+
+        data.update(
+            dict(
+                net_arch=self.net_args["net_arch"],
+                activation_fn=self.net_args["activation_fn"],
+                lr_schedule=self._dummy_schedule,  # dummy lr schedule, not needed for loading policy alone
+                optimizer_class=self.optimizer_class,
+                optimizer_kwargs=self.optimizer_kwargs,
+                features_extractor_class=self.features_extractor_class,
+                features_extractor_kwargs=self.features_extractor_kwargs,
+            )
+        )
+        return data
+
+
+MlpPolicy = DQNPolicy
+
+
+class CnnPolicy(DQNPolicy):
+    """
+    Policy class for DQN when using images as input.
+
+    :param observation_space: Observation space
+    :param action_space: Action space
+    :param lr_schedule: Learning rate schedule (could be constant)
+    :param net_arch: The specification of the policy and value networks.
+    :param activation_fn: Activation function
+    :param features_extractor_class: Features extractor to use.
+    :param normalize_images: Whether to normalize images or not,
+         dividing by 255.0 (True by default)
+    :param optimizer_class: The optimizer to use,
+        ``th.optim.Adam`` by default
+    :param optimizer_kwargs: Additional keyword arguments,
+        excluding the learning rate, to pass to the optimizer
+    """
+
+    def __init__(
+        self,
+        observation_space: gym.spaces.Space,
+        action_space: gym.spaces.Space,
+        lr_schedule: Schedule,
+        net_arch: Optional[List[int]] = None,
+        activation_fn: Type[nn.Module] = nn.ReLU,
+        features_extractor_class: Type[BaseFeaturesExtractor] = NatureCNN,
+        features_extractor_kwargs: Optional[Dict[str, Any]] = None,
+        normalize_images: bool = True,
+        optimizer_class: Type[th.optim.Optimizer] = th.optim.Adam,
+        optimizer_kwargs: Optional[Dict[str, Any]] = None,
+    ):
+        super(CnnPolicy, self).__init__(
+            observation_space,
+            action_space,
+            lr_schedule,
+            net_arch,
+            activation_fn,
+            features_extractor_class,
+            features_extractor_kwargs,
+            normalize_images,
+            optimizer_class,
+            optimizer_kwargs,
+        )
+
+
+register_policy("MlpPolicy", MlpPolicy)
+register_policy("CnnPolicy", CnnPolicy)

crazy_functions/test_project/python/dqn/来源

@@ -0,0 +1,2 @@
+github stablebaseline3
+https://github.com/DLR-RM/stable-baselines3

crazy_functions/test_project/其他测试

@@ -0,0 +1,27 @@
+"In practice, we found that a high-entropy initial state is more likely to increase the speed of training.
+The entropy is calculated by:
+$$H=-\sum_{k= 1}^{n_k} p(k) \cdot \log p(k), p(k)=\frac{|A_k|}{|\mathcal{A}|}$$
+where $H$ is the entropy, $|A_k|$ is the number of agent nodes in $k$-th cluster, $|\mathcal{A}|$ is the total number of agents.
+To ensure the Cooperation Graph initialization has higher entropy,
+we will randomly generate multiple initial states, 
+rank by their entropy and then pick the one with maximum $H$."
+
+```
+FROM ubuntu:latest
+
+RUN apt-get update && \
+    apt-get install -y python3 python3-pip && \
+    rm -rf /var/lib/apt/lists/*
+
+RUN echo '[global]' > /etc/pip.conf && \
+    echo 'index-url = https://mirrors.aliyun.com/pypi/simple/' >> /etc/pip.conf && \
+    echo 'trusted-host = mirrors.aliyun.com' >> /etc/pip.conf
+
+RUN pip3 install gradio requests[socks] mdtex2html
+
+COPY . /gpt
+WORKDIR /gpt
+
+
+CMD ["python3", "main.py"]
+```

crazy_functions/下载arxiv论文翻译摘要.py

@@ -0,0 +1,194 @@
+from toolbox import update_ui
+from toolbox import CatchException, report_execption, write_results_to_file, get_conf
+import re, requests, unicodedata, os
+from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
+def download_arxiv_(url_pdf):
+    if 'arxiv.org' not in url_pdf:
+        if ('.' in url_pdf) and ('/' not in url_pdf):
+            new_url = 'https://arxiv.org/abs/'+url_pdf
+            print('下载编号：', url_pdf, '自动定位：', new_url)
+            # download_arxiv_(new_url)
+            return download_arxiv_(new_url)
+        else:
+            print('不能识别的URL！')
+            return None
+    if 'abs' in url_pdf:
+        url_pdf = url_pdf.replace('abs', 'pdf')
+        url_pdf = url_pdf + '.pdf'
+
+    url_abs = url_pdf.replace('.pdf', '').replace('pdf', 'abs')
+    title, other_info = get_name(_url_=url_abs)
+
+    paper_id = title.split()[0]  # '[1712.00559]'
+    if '2' in other_info['year']:
+        title = other_info['year'] + ' ' + title
+
+    known_conf = ['NeurIPS', 'NIPS', 'Nature', 'Science', 'ICLR', 'AAAI']
+    for k in known_conf:
+        if k in other_info['comment']:
+            title = k + ' ' + title
+
+    download_dir = './gpt_log/arxiv/'
+    os.makedirs(download_dir, exist_ok=True)
+
+    title_str = title.replace('?', '？')\
+        .replace(':', '：')\
+        .replace('\"', '“')\
+        .replace('\n', '')\
+        .replace('  ', ' ')\
+        .replace('  ', ' ')
+
+    requests_pdf_url = url_pdf
+    file_path = download_dir+title_str
+    # if os.path.exists(file_path):
+    #     print('返回缓存文件')
+    #     return './gpt_log/arxiv/'+title_str
+
+    print('下载中')
+    proxies, = get_conf('proxies')
+    r = requests.get(requests_pdf_url, proxies=proxies)
+    with open(file_path, 'wb+') as f:
+        f.write(r.content)
+    print('下载完成')
+
+    # print('输出下载命令：','aria2c -o \"%s\" %s'%(title_str,url_pdf))
+    # subprocess.call('aria2c --all-proxy=\"172.18.116.150:11084\" -o \"%s\" %s'%(download_dir+title_str,url_pdf), shell=True)
+
+    x = "%s  %s %s.bib" % (paper_id, other_info['year'], other_info['authors'])
+    x = x.replace('?', '？')\
+        .replace(':', '：')\
+        .replace('\"', '“')\
+        .replace('\n', '')\
+        .replace('  ', ' ')\
+        .replace('  ', ' ')
+    return './gpt_log/arxiv/'+title_str, other_info
+
+
+def get_name(_url_):
+    import os
+    from bs4 import BeautifulSoup
+    print('正在获取文献名！')
+    print(_url_)
+
+    # arxiv_recall = {}
+    # if os.path.exists('./arxiv_recall.pkl'):
+    #     with open('./arxiv_recall.pkl', 'rb') as f:
+    #         arxiv_recall = pickle.load(f)
+
+    # if _url_ in arxiv_recall:
+    #     print('在缓存中')
+    #     return arxiv_recall[_url_]
+
+    proxies, = get_conf('proxies')
+    res = requests.get(_url_, proxies=proxies)
+
+    bs = BeautifulSoup(res.text, 'html.parser')
+    other_details = {}
+
+    # get year
+    try:
+        year = bs.find_all(class_='dateline')[0].text
+        year = re.search(r'(\d{4})', year, re.M | re.I).group(1)
+        other_details['year'] = year
+        abstract = bs.find_all(class_='abstract mathjax')[0].text
+        other_details['abstract'] = abstract
+    except:
+        other_details['year'] = ''
+        print('年份获取失败')
+
+    # get author
+    try:
+        authors = bs.find_all(class_='authors')[0].text
+        authors = authors.split('Authors:')[1]
+        other_details['authors'] = authors
+    except:
+        other_details['authors'] = ''
+        print('authors获取失败')
+
+    # get comment
+    try:
+        comment = bs.find_all(class_='metatable')[0].text
+        real_comment = None
+        for item in comment.replace('\n', ' ').split('   '):
+            if 'Comments' in item:
+                real_comment = item
+        if real_comment is not None:
+            other_details['comment'] = real_comment
+        else:
+            other_details['comment'] = ''
+    except:
+        other_details['comment'] = ''
+        print('年份获取失败')
+
+    title_str = BeautifulSoup(
+        res.text, 'html.parser').find('title').contents[0]
+    print('获取成功：', title_str)
+    # arxiv_recall[_url_] = (title_str+'.pdf', other_details)
+    # with open('./arxiv_recall.pkl', 'wb') as f:
+    #     pickle.dump(arxiv_recall, f)
+
+    return title_str+'.pdf', other_details
+
+
+
+@CatchException
+def 下载arxiv论文并翻译摘要(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+
+    CRAZY_FUNCTION_INFO = "下载arxiv论文并翻译摘要，函数插件作者[binary-husky]。正在提取摘要并下载PDF文档……"
+    import glob
+    import os
+
+    # 基本信息：功能、贡献者
+    chatbot.append(["函数插件功能？", CRAZY_FUNCTION_INFO])
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import pdfminer, bs4
+    except:
+        report_execption(chatbot, history, 
+            a = f"解析项目: {txt}", 
+            b = f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade pdfminer beautifulsoup4```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 清空历史，以免输入溢出
+    history = []
+
+    # 提取摘要，下载PDF文档
+    try:
+        pdf_path, info = download_arxiv_(txt)
+    except:
+        report_execption(chatbot, history, 
+            a = f"解析项目: {txt}", 
+            b = f"下载pdf文件未成功")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    
+    # 翻译摘要等
+    i_say =            f"请你阅读以下学术论文相关的材料，提取摘要，翻译为中文。材料如下：{str(info)}"
+    i_say_show_user =  f'请你阅读以下学术论文相关的材料，提取摘要，翻译为中文。论文：{pdf_path}'
+    chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+    msg = '正常'
+    # ** gpt request **
+    # 单线，获取文章meta信息
+    gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
+        inputs=i_say,
+        inputs_show_user=i_say_show_user,
+        llm_kwargs=llm_kwargs,
+        chatbot=chatbot, history=[],
+        sys_prompt="Your job is to collect information from materials and translate to Chinese。",
+    )
+
+    chatbot[-1] = (i_say_show_user, gpt_say)
+    history.append(i_say_show_user); history.append(gpt_say)
+    yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+    # 写入文件
+    import shutil
+    # 重置文件的创建时间
+    shutil.copyfile(pdf_path, f'./gpt_log/{os.path.basename(pdf_path)}'); os.remove(pdf_path)
+    res = write_results_to_file(history)
+    chatbot.append(("完成了吗？", res + "\n\nPDF文件也已经下载"))
+    yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+

crazy_functions/代码重写为全英文_多线程.py

@@ -0,0 +1,139 @@
+import threading
+from request_llm.bridge_chatgpt import predict_no_ui_long_connection
+from toolbox import update_ui
+from toolbox import CatchException, write_results_to_file, report_execption
+from .crazy_utils import breakdown_txt_to_satisfy_token_limit
+
+def extract_code_block_carefully(txt):
+    splitted = txt.split('```')
+    n_code_block_seg = len(splitted) - 1
+    if n_code_block_seg <= 1: return txt
+    # 剩下的情况都开头除去 ``` 结尾除去一次 ```
+    txt_out = '```'.join(splitted[1:-1])
+    return txt_out
+
+
+
+def break_txt_into_half_at_some_linebreak(txt):
+    lines = txt.split('\n')
+    n_lines = len(lines)
+    pre = lines[:(n_lines//2)]
+    post = lines[(n_lines//2):]
+    return "\n".join(pre), "\n".join(post)
+
+
+@CatchException
+def 全项目切换英文(txt, llm_kwargs, plugin_kwargs, chatbot, history, sys_prompt, web_port):
+    # 第1步：清空历史，以免输入溢出
+    history = []
+
+    # 第2步：尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import tiktoken
+    except:
+        report_execption(chatbot, history, 
+            a = f"解析项目: {txt}", 
+            b = f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade tiktoken```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 第3步：集合文件
+    import time, glob, os, shutil, re
+    os.makedirs('gpt_log/generated_english_version', exist_ok=True)
+    os.makedirs('gpt_log/generated_english_version/crazy_functions', exist_ok=True)
+    file_manifest = [f for f in glob.glob('./*.py') if ('test_project' not in f) and ('gpt_log' not in f)] + \
+                    [f for f in glob.glob('./crazy_functions/*.py') if ('test_project' not in f) and ('gpt_log' not in f)]
+    # file_manifest = ['./toolbox.py']
+    i_say_show_user_buffer = []
+
+    # 第4步：随便显示点什么防止卡顿的感觉
+    for index, fp in enumerate(file_manifest):
+        # if 'test_project' in fp: continue
+        with open(fp, 'r', encoding='utf-8', errors='replace') as f:
+            file_content = f.read()
+        i_say_show_user =f'[{index}/{len(file_manifest)}] 接下来请将以下代码中包含的所有中文转化为英文，只输出转化后的英文代码，请用代码块输出代码: {os.path.abspath(fp)}'
+        i_say_show_user_buffer.append(i_say_show_user)
+        chatbot.append((i_say_show_user, "[Local Message] 等待多线程操作，中间过程不予显示."))
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+
+    # 第5步：Token限制下的截断与处理
+    MAX_TOKEN = 3000
+    import tiktoken
+    from toolbox import get_conf
+    enc = tiktoken.encoding_for_model(*get_conf('LLM_MODEL'))
+    def get_token_fn(txt): return len(enc.encode(txt))
+
+
+    # 第6步：任务函数
+    mutable_return = [None for _ in file_manifest]
+    observe_window = [[""] for _ in file_manifest]
+    def thread_worker(fp,index):
+        if index > 10: 
+            time.sleep(60)
+            print('Openai 限制免费用户每分钟20次请求，降低请求频率中。')
+        with open(fp, 'r', encoding='utf-8', errors='replace') as f:
+            file_content = f.read()
+        i_say_template = lambda fp, file_content: f'接下来请将以下代码中包含的所有中文转化为英文，只输出代码，文件名是{fp}，文件代码是 ```{file_content}```'
+        try:
+            gpt_say = ""
+            # 分解代码文件
+            file_content_breakdown = breakdown_txt_to_satisfy_token_limit(file_content, get_token_fn, MAX_TOKEN)
+            for file_content_partial in file_content_breakdown:
+                i_say = i_say_template(fp, file_content_partial)
+                # # ** gpt request **
+                gpt_say_partial = predict_no_ui_long_connection(inputs=i_say, llm_kwargs=llm_kwargs, history=[], sys_prompt=sys_prompt, observe_window=observe_window[index])
+                gpt_say_partial = extract_code_block_carefully(gpt_say_partial)
+                gpt_say += gpt_say_partial
+            mutable_return[index] = gpt_say
+        except ConnectionAbortedError as token_exceed_err:
+            print('至少一个线程任务Token溢出而失败', e)
+        except Exception as e:
+            print('至少一个线程任务意外失败', e)
+
+    # 第7步：所有线程同时开始执行任务函数
+    handles = [threading.Thread(target=thread_worker, args=(fp,index)) for index, fp in enumerate(file_manifest)]
+    for h in handles:
+        h.daemon = True
+        h.start()
+    chatbot.append(('开始了吗？', f'多线程操作已经开始'))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 第8步：循环轮询各个线程是否执行完毕
+    cnt = 0
+    while True:
+        cnt += 1
+        time.sleep(0.2)
+        th_alive = [h.is_alive() for h in handles]
+        if not any(th_alive): break
+        # 更好的UI视觉效果
+        observe_win = []
+        for thread_index, alive in enumerate(th_alive): 
+            observe_win.append("[ ..."+observe_window[thread_index][0][-60:].replace('\n','').replace('```','...').replace(' ','.').replace('<br/>','.....').replace('$','.')+"... ]")
+        stat = [f'执行中: {obs}\n\n' if alive else '已完成\n\n' for alive, obs in zip(th_alive, observe_win)]
+        stat_str = ''.join(stat)
+        chatbot[-1] = (chatbot[-1][0], f'多线程操作已经开始，完成情况: \n\n{stat_str}' + ''.join(['.']*(cnt%10+1)))
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 第9步：把结果写入文件
+    for index, h in enumerate(handles):
+        h.join() # 这里其实不需要join了，肯定已经都结束了
+        fp = file_manifest[index]
+        gpt_say = mutable_return[index]
+        i_say_show_user = i_say_show_user_buffer[index]
+
+        where_to_relocate = f'gpt_log/generated_english_version/{fp}'
+        if gpt_say is not None:
+            with open(where_to_relocate, 'w+', encoding='utf-8') as f:  
+                f.write(gpt_say)
+        else:  # 失败
+            shutil.copyfile(file_manifest[index], where_to_relocate)
+        chatbot.append((i_say_show_user, f'[Local Message] 已完成{os.path.abspath(fp)}的转化，\n\n存入{os.path.abspath(where_to_relocate)}'))
+        history.append(i_say_show_user); history.append(gpt_say)
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        time.sleep(1)
+
+    # 第10步：备份一个文件
+    res = write_results_to_file(history)
+    chatbot.append(("生成一份任务执行报告", res))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面

crazy_functions/总结word文档.py

@@ -0,0 +1,139 @@
+from toolbox import update_ui
+from toolbox import CatchException, report_execption, write_results_to_file
+from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
+fast_debug = False
+
+
+def 解析docx(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt):
+    import time, os
+    # pip install python-docx 用于docx格式，跨平台
+    # pip install pywin32 用于doc格式，仅支持Win平台
+
+    print('begin analysis on:', file_manifest)
+    for index, fp in enumerate(file_manifest):
+        if fp.split(".")[-1] == "docx":
+            from docx import Document
+            doc = Document(fp)
+            file_content = "\n".join([para.text for para in doc.paragraphs])
+        else:
+            import win32com.client
+            word = win32com.client.Dispatch("Word.Application")
+            word.visible = False
+            # 打开文件
+            print('fp', os.getcwd())
+            doc = word.Documents.Open(os.getcwd() + '/' + fp)
+            # file_content = doc.Content.Text
+            doc = word.ActiveDocument
+            file_content = doc.Range().Text
+            doc.Close()
+            word.Quit()
+
+        print(file_content)
+
+        prefix = "接下来请你逐文件分析下面的论文文件，" if index == 0 else ""
+        # private_upload里面的文件名在解压zip后容易出现乱码（rar和7z格式正常），故可以只分析文章内容，不输入文件名
+        i_say = prefix + f'请对下面的文章片段用中英文做概述，文件名是{os.path.relpath(fp, project_folder)},' \
+                         f'文章内容是 ```{file_content}```'
+        i_say_show_user = prefix + f'[{index+1}/{len(file_manifest)}] 假设你是论文审稿专家，请对下面的文章片段做概述: {os.path.abspath(fp)}'
+        chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+        if not fast_debug:
+            msg = '正常'
+            # ** gpt request **
+            gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
+                inputs=i_say, 
+                inputs_show_user=i_say_show_user, 
+                llm_kwargs=llm_kwargs,
+                chatbot=chatbot, 
+                history=[],
+                sys_prompt="总结文章。"
+            )  # 带超时倒计时
+            chatbot[-1] = (i_say_show_user, gpt_say)
+            history.append(i_say_show_user)
+            history.append(gpt_say)
+            yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+            if not fast_debug: time.sleep(2)
+
+    """
+    # 可按需启用
+    i_say = f'根据你上述的分析，对全文进行概括，用学术性语言写一段中文摘要，然后再写一篇英文的。'
+    chatbot.append((i_say, "[Local Message] waiting gpt response."))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+
+    i_say = f'我想让你做一个论文写作导师。您的任务是使用人工智能工具（例如自然语言处理）提供有关如何改进其上述文章的反馈。' \
+            f'您还应该利用您在有效写作技巧方面的修辞知识和经验来建议作者可以更好地以书面形式表达他们的想法和想法的方法。' \
+            f'根据你之前的分析，提出建议'
+    chatbot.append((i_say, "[Local Message] waiting gpt response."))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+  
+    """
+
+    if not fast_debug:
+        msg = '正常'
+        # ** gpt request **
+        gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
+            inputs=i_say, 
+            inputs_show_user=i_say, 
+            llm_kwargs=llm_kwargs,
+            chatbot=chatbot, 
+            history=history,
+            sys_prompt="总结文章。"
+        )  # 带超时倒计时
+        chatbot[-1] = (i_say, gpt_say)
+        history.append(i_say)
+        history.append(gpt_say)
+        yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+        res = write_results_to_file(history)
+        chatbot.append(("完成了吗？", res))
+        yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+
+
+@CatchException
+def 总结word文档(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    import glob, os
+
+    # 基本信息：功能、贡献者
+    chatbot.append([
+        "函数插件功能？",
+        "批量总结Word文档。函数插件贡献者: JasonGuo1"])
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        from docx import Document
+    except:
+        report_execption(chatbot, history,
+                         a=f"解析项目: {txt}",
+                         b=f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade python-docx pywin32```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 清空历史，以免输入溢出
+    history = []
+
+    # 检测输入参数，如没有给定输入参数，直接退出
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a=f"解析项目: {txt}", b=f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 搜索需要处理的文件清单
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.docx', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.doc', recursive=True)]
+    # [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)] + \
+    # [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)] + \
+    # [f for f in glob.glob(f'{project_folder}/**/*.c', recursive=True)]
+
+    # 如果没找到任何文件
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a=f"解析项目: {txt}", b=f"找不到任何.docx或doc文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 开始正式执行任务
+    yield from 解析docx(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)

crazy_functions/批量总结PDF文档.py

@@ -0,0 +1,166 @@
+from toolbox import update_ui
+from toolbox import CatchException, report_execption, write_results_to_file
+import re
+import unicodedata
+fast_debug = False
+from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
+
+def is_paragraph_break(match):
+    """
+    根据给定的匹配结果来判断换行符是否表示段落分隔。
+    如果换行符前为句子结束标志（句号，感叹号，问号），且下一个字符为大写字母，则换行符更有可能表示段落分隔。
+    也可以根据之前的内容长度来判断段落是否已经足够长。
+    """
+    prev_char, next_char = match.groups()
+
+    # 句子结束标志
+    sentence_endings = ".!?"
+
+    # 设定一个最小段落长度阈值
+    min_paragraph_length = 140
+
+    if prev_char in sentence_endings and next_char.isupper() and len(match.string[:match.start(1)]) > min_paragraph_length:
+        return "\n\n" 
+    else:
+        return " "
+
+def normalize_text(text):
+    """
+    通过把连字（ligatures）等文本特殊符号转换为其基本形式来对文本进行归一化处理。
+    例如，将连字 "fi" 转换为 "f" 和 "i"。
+    """
+    # 对文本进行归一化处理，分解连字
+    normalized_text = unicodedata.normalize("NFKD", text)
+
+    # 替换其他特殊字符
+    cleaned_text = re.sub(r'[^\x00-\x7F]+', '', normalized_text)
+
+    return cleaned_text
+
+def clean_text(raw_text):
+    """
+    对从 PDF 提取出的原始文本进行清洗和格式化处理。
+    1. 对原始文本进行归一化处理。
+    2. 替换跨行的连词，例如 “Espe-\ncially” 转换为 “Especially”。
+    3. 根据 heuristic 规则判断换行符是否是段落分隔，并相应地进行替换。
+    """
+    # 对文本进行归一化处理
+    normalized_text = normalize_text(raw_text)
+
+    # 替换跨行的连词
+    text = re.sub(r'(\w+-\n\w+)', lambda m: m.group(1).replace('-\n', ''), normalized_text)
+
+    # 根据前后相邻字符的特点，找到原文本中的换行符
+    newlines = re.compile(r'(\S)\n(\S)')
+
+    # 根据 heuristic 规则，用空格或段落分隔符替换原换行符
+    final_text = re.sub(newlines, lambda m: m.group(1) + is_paragraph_break(m) + m.group(2), text)
+
+    return final_text.strip()
+
+def 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt):
+    import time, glob, os, fitz
+    print('begin analysis on:', file_manifest)
+    for index, fp in enumerate(file_manifest):
+        with fitz.open(fp) as doc:
+            file_content = ""
+            for page in doc:
+                file_content += page.get_text()
+            file_content = clean_text(file_content)
+            print(file_content)
+
+        prefix = "接下来请你逐文件分析下面的论文文件，概括其内容" if index==0 else ""
+        i_say = prefix + f'请对下面的文章片段用中文做一个概述，文件名是{os.path.relpath(fp, project_folder)}，文章内容是 ```{file_content}```'
+        i_say_show_user = prefix + f'[{index}/{len(file_manifest)}] 请对下面的文章片段做一个概述: {os.path.abspath(fp)}'
+        chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+        if not fast_debug: 
+            msg = '正常'
+            # ** gpt request **
+            gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
+                inputs=i_say, 
+                inputs_show_user=i_say_show_user, 
+                llm_kwargs=llm_kwargs,
+                chatbot=chatbot, 
+                history=[],
+                sys_prompt="总结文章。"
+            )  # 带超时倒计时
+                
+
+            chatbot[-1] = (i_say_show_user, gpt_say)
+            history.append(i_say_show_user); history.append(gpt_say)
+            yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+            if not fast_debug: time.sleep(2)
+
+    all_file = ', '.join([os.path.relpath(fp, project_folder) for index, fp in enumerate(file_manifest)])
+    i_say = f'根据以上你自己的分析，对全文进行概括，用学术性语言写一段中文摘要，然后再写一段英文摘要（包括{all_file}）。'
+    chatbot.append((i_say, "[Local Message] waiting gpt response."))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    if not fast_debug: 
+        msg = '正常'
+        # ** gpt request **
+        gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
+            inputs=i_say, 
+            inputs_show_user=i_say, 
+            llm_kwargs=llm_kwargs,
+            chatbot=chatbot, 
+            history=history,
+            sys_prompt="总结文章。"
+        )  # 带超时倒计时
+
+        chatbot[-1] = (i_say, gpt_say)
+        history.append(i_say); history.append(gpt_say)
+        yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+        res = write_results_to_file(history)
+        chatbot.append(("完成了吗？", res))
+        yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+
+
+@CatchException
+def 批量总结PDF文档(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    import glob, os
+
+    # 基本信息：功能、贡献者
+    chatbot.append([
+        "函数插件功能？",
+        "批量总结PDF文档。函数插件贡献者: ValeriaWong，Eralien"])
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import fitz
+    except:
+        report_execption(chatbot, history, 
+            a = f"解析项目: {txt}", 
+            b = f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade pymupdf```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 清空历史，以免输入溢出
+    history = []
+
+    # 检测输入参数，如没有给定输入参数，直接退出
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 搜索需要处理的文件清单
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.pdf', recursive=True)] # + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)] + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)] + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.c', recursive=True)]
+    
+    # 如果没找到任何文件
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex或.pdf文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 开始正式执行任务
+    yield from 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)

crazy_functions/批量总结PDF文档pdfminer.py

@@ -0,0 +1,160 @@
+from toolbox import update_ui
+from toolbox import CatchException, report_execption, write_results_to_file
+from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
+
+fast_debug = False
+
+def readPdf(pdfPath):
+    """
+    读取pdf文件，返回文本内容
+    """
+    import pdfminer
+    from pdfminer.pdfparser import PDFParser
+    from pdfminer.pdfdocument import PDFDocument
+    from pdfminer.pdfpage import PDFPage, PDFTextExtractionNotAllowed
+    from pdfminer.pdfinterp import PDFResourceManager, PDFPageInterpreter
+    from pdfminer.pdfdevice import PDFDevice
+    from pdfminer.layout import LAParams
+    from pdfminer.converter import PDFPageAggregator
+
+    fp = open(pdfPath, 'rb')
+
+    # Create a PDF parser object associated with the file object
+    parser = PDFParser(fp)
+
+    # Create a PDF document object that stores the document structure.
+    # Password for initialization as 2nd parameter
+    document = PDFDocument(parser)
+    # Check if the document allows text extraction. If not, abort.
+    if not document.is_extractable:
+        raise PDFTextExtractionNotAllowed
+
+    # Create a PDF resource manager object that stores shared resources.
+    rsrcmgr = PDFResourceManager()
+
+    # Create a PDF device object.
+    # device = PDFDevice(rsrcmgr)
+
+    # BEGIN LAYOUT ANALYSIS.
+    # Set parameters for analysis.
+    laparams = LAParams(
+        char_margin=10.0,
+        line_margin=0.2,
+        boxes_flow=0.2,
+        all_texts=False,
+    )
+    # Create a PDF page aggregator object.
+    device = PDFPageAggregator(rsrcmgr, laparams=laparams)
+    # Create a PDF interpreter object.
+    interpreter = PDFPageInterpreter(rsrcmgr, device)
+
+    # loop over all pages in the document
+    outTextList = []
+    for page in PDFPage.create_pages(document):
+        # read the page into a layout object
+        interpreter.process_page(page)
+        layout = device.get_result()
+        for obj in layout._objs:
+            if isinstance(obj, pdfminer.layout.LTTextBoxHorizontal):
+                # print(obj.get_text())
+                outTextList.append(obj.get_text())
+
+    return outTextList
+
+
+def 解析Paper(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt):
+    import time, glob, os
+    from bs4 import BeautifulSoup
+    print('begin analysis on:', file_manifest)
+    for index, fp in enumerate(file_manifest):
+        if ".tex" in fp:
+            with open(fp, 'r', encoding='utf-8', errors='replace') as f:
+                file_content = f.read()
+        if ".pdf" in fp.lower():
+            file_content = readPdf(fp)
+            file_content = BeautifulSoup(''.join(file_content), features="lxml").body.text.encode('gbk', 'ignore').decode('gbk')
+
+        prefix = "接下来请你逐文件分析下面的论文文件，概括其内容" if index==0 else ""
+        i_say = prefix + f'请对下面的文章片段用中文做一个概述，文件名是{os.path.relpath(fp, project_folder)}，文章内容是 ```{file_content}```'
+        i_say_show_user = prefix + f'[{index}/{len(file_manifest)}] 请对下面的文章片段做一个概述: {os.path.abspath(fp)}'
+        chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+        if not fast_debug:
+            msg = '正常'
+            # ** gpt request **
+            gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
+                inputs=i_say, 
+                inputs_show_user=i_say_show_user, 
+                llm_kwargs=llm_kwargs,
+                chatbot=chatbot, 
+                history=[],
+                sys_prompt="总结文章。"
+            )  # 带超时倒计时
+            chatbot[-1] = (i_say_show_user, gpt_say)
+            history.append(i_say_show_user); history.append(gpt_say)
+            yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+            if not fast_debug: time.sleep(2)
+
+    all_file = ', '.join([os.path.relpath(fp, project_folder) for index, fp in enumerate(file_manifest)])
+    i_say = f'根据以上你自己的分析，对全文进行概括，用学术性语言写一段中文摘要，然后再写一段英文摘要（包括{all_file}）。'
+    chatbot.append((i_say, "[Local Message] waiting gpt response."))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    if not fast_debug:
+        msg = '正常'
+        # ** gpt request **
+        gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
+            inputs=i_say, 
+            inputs_show_user=i_say, 
+            llm_kwargs=llm_kwargs,
+            chatbot=chatbot, 
+            history=history,
+            sys_prompt="总结文章。"
+        )  # 带超时倒计时
+        chatbot[-1] = (i_say, gpt_say)
+        history.append(i_say); history.append(gpt_say)
+        yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+        res = write_results_to_file(history)
+        chatbot.append(("完成了吗？", res))
+        yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+
+
+
+@CatchException
+def 批量总结PDF文档pdfminer(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+
+    # 基本信息：功能、贡献者
+    chatbot.append([
+        "函数插件功能？",
+        "批量总结PDF文档，此版本使用pdfminer插件，带token约简功能。函数插件贡献者: Euclid-Jie。"])
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import pdfminer, bs4
+    except:
+        report_execption(chatbot, history, 
+            a = f"解析项目: {txt}", 
+            b = f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade pdfminer beautifulsoup4```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.pdf', recursive=True)] # + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)] + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.c', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex或pdf文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 解析Paper(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)
+

crazy_functions/批量翻译PDF文档_多线程.py

@@ -0,0 +1,131 @@
+from toolbox import CatchException, report_execption, write_results_to_file
+from toolbox import update_ui
+from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
+from .crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency
+from .crazy_utils import read_and_clean_pdf_text
+from colorful import *
+
+@CatchException
+def 批量翻译PDF文档(txt, llm_kwargs, plugin_kwargs, chatbot, history, sys_prompt, web_port):
+    import glob
+    import os
+
+    # 基本信息：功能、贡献者
+    chatbot.append([
+        "函数插件功能？",
+        "批量总结PDF文档。函数插件贡献者: Binary-Husky"])
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import fitz
+        import tiktoken
+    except:
+        report_execption(chatbot, history,
+                         a=f"解析项目: {txt}",
+                         b=f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade pymupdf tiktoken```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 清空历史，以免输入溢出
+    history = []
+
+    # 检测输入参数，如没有给定输入参数，直接退出
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "":
+            txt = '空空如也的输入栏'
+        report_execption(chatbot, history,
+                         a=f"解析项目: {txt}", b=f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 搜索需要处理的文件清单
+    file_manifest = [f for f in glob.glob(
+        f'{project_folder}/**/*.pdf', recursive=True)]
+
+    # 如果没找到任何文件
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history,
+                         a=f"解析项目: {txt}", b=f"找不到任何.tex或.pdf文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 开始正式执行任务
+    yield from 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, sys_prompt)
+
+
+def 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, sys_prompt):
+    import os
+    import tiktoken
+    TOKEN_LIMIT_PER_FRAGMENT = 1600
+    generated_conclusion_files = []
+    for index, fp in enumerate(file_manifest):
+
+        # 读取PDF文件
+        file_content, page_one = read_and_clean_pdf_text(fp)
+
+        # 递归地切割PDF文件
+        from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
+        from toolbox import get_conf
+        enc = tiktoken.encoding_for_model(*get_conf('LLM_MODEL'))
+        def get_token_num(txt): return len(enc.encode(txt))
+        paper_fragments = breakdown_txt_to_satisfy_token_limit_for_pdf(
+            txt=file_content,  get_token_fn=get_token_num, limit=TOKEN_LIMIT_PER_FRAGMENT)
+        page_one_fragments = breakdown_txt_to_satisfy_token_limit_for_pdf(
+            txt=str(page_one), get_token_fn=get_token_num, limit=TOKEN_LIMIT_PER_FRAGMENT//4)
+
+        # 为了更好的效果，我们剥离Introduction之后的部分（如果有）
+        paper_meta = page_one_fragments[0].split('introduction')[0].split('Introduction')[0].split('INTRODUCTION')[0]
+        
+        # 单线，获取文章meta信息
+        paper_meta_info = yield from request_gpt_model_in_new_thread_with_ui_alive(
+            inputs=f"以下是一篇学术论文的基础信息，请从中提取出“标题”、“收录会议或期刊”、“作者”、“摘要”、“编号”、“作者邮箱”这六个部分。请用markdown格式输出，最后用中文翻译摘要部分。请提取：{paper_meta}",
+            inputs_show_user=f"请从{fp}中提取出“标题”、“收录会议或期刊”等基本信息。",
+            llm_kwargs=llm_kwargs,
+            chatbot=chatbot, history=[],
+            sys_prompt="Your job is to collect information from materials。",
+        )
+
+        # 多线，翻译
+        gpt_response_collection = yield from request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
+            inputs_array=[
+                f"以下是你需要翻译的论文片段：\n{frag}" for frag in paper_fragments],
+            inputs_show_user_array=[f"\n---\n 原文： \n\n {frag.replace('#', '')}  \n---\n 翻译：\n " for frag in paper_fragments],
+            llm_kwargs=llm_kwargs,
+            chatbot=chatbot,
+            history_array=[[paper_meta] for _ in paper_fragments],
+            sys_prompt_array=[
+                "请你作为一个学术翻译，负责把学术论文的片段准确翻译成中文。" for _ in paper_fragments],
+            max_workers=16  # OpenAI所允许的最大并行过载
+        )
+
+        # 整理报告的格式
+        for i,k in enumerate(gpt_response_collection): 
+            if i%2==0:
+                gpt_response_collection[i] = f"\n\n---\n\n ## 原文[{i//2}/{len(gpt_response_collection)//2}]： \n\n {paper_fragments[i//2].replace('#', '')}  \n\n---\n\n ## 翻译[{i//2}/{len(gpt_response_collection)//2}]：\n "
+            else:
+                gpt_response_collection[i] = gpt_response_collection[i]
+        final = ["一、论文概况\n\n---\n\n", paper_meta_info.replace('# ', '### ') + '\n\n---\n\n', "二、论文翻译", ""]
+        final.extend(gpt_response_collection)
+        create_report_file_name = f"{os.path.basename(fp)}.trans.md"
+        res = write_results_to_file(final, file_name=create_report_file_name)
+
+        # 更新UI
+        generated_conclusion_files.append(f'./gpt_log/{create_report_file_name}')
+        chatbot.append((f"{fp}完成了吗？", res))
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 准备文件的下载
+    import shutil
+    for pdf_path in generated_conclusion_files:
+        # 重命名文件
+        rename_file = f'./gpt_log/总结论文-{os.path.basename(pdf_path)}'
+        if os.path.exists(rename_file):
+            os.remove(rename_file)
+        shutil.copyfile(pdf_path, rename_file)
+        if os.path.exists(pdf_path):
+            os.remove(pdf_path)
+    chatbot.append(("给出输出文件清单", str(generated_conclusion_files)))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面

crazy_functions/理解PDF文档内容.py

@@ -0,0 +1,111 @@
+from toolbox import update_ui
+from toolbox import CatchException, report_execption
+from .crazy_utils import read_and_clean_pdf_text
+from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
+fast_debug = False
+
+
+def 解析PDF(file_name, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt):
+    import tiktoken
+    print('begin analysis on:', file_name)
+    file_content, page_one = read_and_clean_pdf_text(file_name)
+
+    ############################## <第零步，从摘要中提取高价值信息，放到history中> ##################################
+    # 递归地切割PDF文件，每一块（尽量是完整的一个section，比如introduction，experiment等，必要时再进行切割）
+    # 的长度必须小于 2500 个 Token
+    TOKEN_LIMIT_PER_FRAGMENT = 2500
+
+    from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
+    from toolbox import get_conf
+    enc = tiktoken.encoding_for_model(*get_conf('LLM_MODEL'))
+    def get_token_num(txt): return len(enc.encode(txt))
+    paper_fragments = breakdown_txt_to_satisfy_token_limit_for_pdf(
+        txt=file_content,  get_token_fn=get_token_num, limit=TOKEN_LIMIT_PER_FRAGMENT)
+    page_one_fragments = breakdown_txt_to_satisfy_token_limit_for_pdf(
+        txt=str(page_one), get_token_fn=get_token_num, limit=TOKEN_LIMIT_PER_FRAGMENT//4)
+    # 为了更好的效果，我们剥离Introduction之后的部分（如果有）
+    paper_meta = page_one_fragments[0].split('introduction')[0].split('Introduction')[0].split('INTRODUCTION')[0]
+    
+    ############################## <第一步，从摘要中提取高价值信息，放到history中> ##################################
+    final_results = []
+    final_results.append(paper_meta)
+
+    ############################## <第二步，迭代地历遍整个文章，提取精炼信息> ##################################
+    i_say_show_user = f'首先你在英文语境下通读整篇论文。'; gpt_say = "[Local Message] 收到。"           # 用户提示
+    chatbot.append([i_say_show_user, gpt_say]); yield from update_ui(chatbot=chatbot, history=[])    # 更新UI
+
+    iteration_results = []
+    last_iteration_result = paper_meta  # 初始值是摘要
+    MAX_WORD_TOTAL = 4096
+    n_fragment = len(paper_fragments)
+    if n_fragment >= 20: print('文章极长，不能达到预期效果')
+    for i in range(n_fragment):
+        NUM_OF_WORD = MAX_WORD_TOTAL // n_fragment
+        i_say = f"Read this section, recapitulate the content of this section with less than {NUM_OF_WORD} words: {paper_fragments[i]}"
+        i_say_show_user = f"[{i+1}/{n_fragment}] Read this section, recapitulate the content of this section with less than {NUM_OF_WORD} words: {paper_fragments[i][:200]}"
+        gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(i_say, i_say_show_user,  # i_say=真正给chatgpt的提问， i_say_show_user=给用户看的提问
+                                                                           llm_kwargs, chatbot, 
+                                                                           history=["The main idea of the previous section is?", last_iteration_result], # 迭代上一次的结果
+                                                                           sys_prompt="Extract the main idea of this section."  # 提示
+                                                                        ) 
+        iteration_results.append(gpt_say)
+        last_iteration_result = gpt_say
+
+    ############################## <第三步，整理history> ##################################
+    final_results.extend(iteration_results)
+    final_results.append(f'接下来，你是一名专业的学术教授，利用以上信息，使用中文回答我的问题。')
+    # 接下来两句话只显示在界面上，不起实际作用
+    i_say_show_user = f'接下来，你是一名专业的学术教授，利用以上信息，使用中文回答我的问题。'; gpt_say = "[Local Message] 收到。"
+    chatbot.append([i_say_show_user, gpt_say])
+
+    ############################## <第四步，设置一个token上限，防止回答时Token溢出> ##################################
+    from .crazy_utils import input_clipping
+    _, final_results = input_clipping("", final_results, max_token_limit=3200)
+    yield from update_ui(chatbot=chatbot, history=final_results) # 注意这里的历史记录被替代了
+
+
+@CatchException
+def 理解PDF文档内容标准文件输入(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    import glob, os
+
+    # 基本信息：功能、贡献者
+    chatbot.append([
+        "函数插件功能？",
+        "理解PDF论文内容，并且将结合上下文内容，进行学术解答。函数插件贡献者: Hanzoe, binary-husky"])
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import fitz
+    except:
+        report_execption(chatbot, history, 
+            a = f"解析项目: {txt}", 
+            b = f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade pymupdf```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 清空历史，以免输入溢出
+    history = []
+
+    # 检测输入参数，如没有给定输入参数，直接退出
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "":
+            txt = '空空如也的输入栏'
+        report_execption(chatbot, history,
+                         a=f"解析项目: {txt}", b=f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 搜索需要处理的文件清单
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.pdf', recursive=True)]
+    # 如果没找到任何文件
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history,
+                         a=f"解析项目: {txt}", b=f"找不到任何.tex或.pdf文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    txt = file_manifest[0]
+    # 开始正式执行任务
+    yield from 解析PDF(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)

crazy_functions/生成函数注释.py

@@ -0,0 +1,54 @@
+from toolbox import update_ui
+from toolbox import CatchException, report_execption, write_results_to_file
+from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
+fast_debug = False
+
+def 生成函数注释(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt):
+    import time, os
+    print('begin analysis on:', file_manifest)
+    for index, fp in enumerate(file_manifest):
+        with open(fp, 'r', encoding='utf-8', errors='replace') as f:
+            file_content = f.read()
+
+        i_say = f'请对下面的程序文件做一个概述，并对文件中的所有函数生成注释，使用markdown表格输出结果，文件名是{os.path.relpath(fp, project_folder)}，文件内容是 ```{file_content}```'
+        i_say_show_user = f'[{index}/{len(file_manifest)}] 请对下面的程序文件做一个概述，并对文件中的所有函数生成注释: {os.path.abspath(fp)}'
+        chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+        if not fast_debug: 
+            msg = '正常'
+            # ** gpt request **
+            gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
+                i_say, i_say_show_user, llm_kwargs, chatbot, history=[], sys_prompt=system_prompt)   # 带超时倒计时
+
+            chatbot[-1] = (i_say_show_user, gpt_say)
+            history.append(i_say_show_user); history.append(gpt_say)
+            yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+            if not fast_debug: time.sleep(2)
+
+    if not fast_debug: 
+        res = write_results_to_file(history)
+        chatbot.append(("完成了吗？", res))
+        yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+
+
+
+@CatchException
+def 批量生成函数注释(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.py', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)]
+
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 生成函数注释(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)

crazy_functions/解析项目源代码.py

@@ -0,0 +1,258 @@
+from toolbox import update_ui
+from toolbox import CatchException, report_execption, write_results_to_file
+
+def 解析源代码新(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt):
+    import os, copy
+    from .crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency
+    from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
+    msg = '正常'
+    inputs_array = []
+    inputs_show_user_array = []
+    history_array = []
+    sys_prompt_array = []
+    report_part_1 = []
+
+    ############################## <第一步，逐个文件分析，多线程> ##################################
+    for index, fp in enumerate(file_manifest):
+        with open(fp, 'r', encoding='utf-8', errors='replace') as f:
+            file_content = f.read()
+        prefix = "接下来请你逐文件分析下面的工程" if index==0 else ""
+        i_say = prefix + f'请对下面的程序文件做一个概述文件名是{os.path.relpath(fp, project_folder)}，文件代码是 ```{file_content}```'
+        i_say_show_user = prefix + f'[{index}/{len(file_manifest)}] 请对下面的程序文件做一个概述: {os.path.abspath(fp)}'
+        # 装载请求内容
+        inputs_array.append(i_say)
+        inputs_show_user_array.append(i_say_show_user)
+        history_array.append([])
+        sys_prompt_array.append("你是一个程序架构分析师，正在分析一个源代码项目。你的回答必须简单明了。")
+
+    gpt_response_collection = yield from request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
+        inputs_array = inputs_array,
+        inputs_show_user_array = inputs_show_user_array,
+        history_array = history_array,
+        sys_prompt_array = sys_prompt_array,
+        llm_kwargs = llm_kwargs,
+        chatbot = chatbot,
+        show_user_at_complete = True
+    )
+
+    report_part_1 = copy.deepcopy(gpt_response_collection)
+    history_to_return = report_part_1
+    res = write_results_to_file(report_part_1)
+    chatbot.append(("完成？", "逐个文件分析已完成。" + res + "\n\n正在开始汇总。"))
+    yield from update_ui(chatbot=chatbot, history=history_to_return) # 刷新界面
+
+    ############################## <存储中间数据进行调试> ##################################
+        
+    # def objdump(obj):
+    #     import pickle
+    #     with open('objdump.tmp', 'wb+') as f:
+    #         pickle.dump(obj, f)
+    #     return
+
+    # def objload():
+    #     import pickle, os
+    #     if not os.path.exists('objdump.tmp'): 
+    #         return
+    #     with open('objdump.tmp', 'rb') as f:
+    #         return pickle.load(f)
+    # objdump([report_part_1, gpt_response_collection, history_to_return, file_manifest, project_folder, fp, llm_kwargs, chatbot])
+    
+    ############################## <第二步，综合，单线程，分组+迭代处理> ##################################
+    batchsize = 16  # 10个文件为一组
+    report_part_2 = []
+    previous_iteration_files = []
+    last_iteration_result = ""
+    while True:
+        if len(file_manifest) == 0: break
+        this_iteration_file_manifest = file_manifest[:batchsize]
+        this_iteration_gpt_response_collection = gpt_response_collection[:batchsize*2]
+        file_rel_path = [os.path.relpath(fp, project_folder) for index, fp in enumerate(this_iteration_file_manifest)]
+        # 把“请对下面的程序文件做一个概述” 替换成 精简的 "文件名：{all_file[index]}"
+        for index, content in enumerate(this_iteration_gpt_response_collection):
+            if index%2==0: this_iteration_gpt_response_collection[index] = f"{file_rel_path[index//2]}" # 只保留文件名节省token
+        previous_iteration_files.extend([os.path.relpath(fp, project_folder) for index, fp in enumerate(this_iteration_file_manifest)])
+        previous_iteration_files_string = ', '.join(previous_iteration_files)
+        current_iteration_focus = ', '.join([os.path.relpath(fp, project_folder) for index, fp in enumerate(this_iteration_file_manifest)])
+        i_say = f'根据以上分析，对程序的整体功能和构架重新做出概括。然后用一张markdown表格整理每个文件的功能（包括{previous_iteration_files_string}）。'
+        inputs_show_user = f'根据以上分析，对程序的整体功能和构架重新做出概括，由于输入长度限制，可能需要分组处理，本组文件为 {current_iteration_focus} + 已经汇总的文件组。'
+        this_iteration_history = copy.deepcopy(this_iteration_gpt_response_collection) 
+        this_iteration_history.append(last_iteration_result)
+        result = yield from request_gpt_model_in_new_thread_with_ui_alive(
+            inputs=i_say, inputs_show_user=inputs_show_user, llm_kwargs=llm_kwargs, chatbot=chatbot, 
+            history=this_iteration_history,   # 迭代之前的分析
+            sys_prompt="你是一个程序架构分析师，正在分析一个项目的源代码。")
+        report_part_2.extend([i_say, result])
+        last_iteration_result = result
+
+        file_manifest = file_manifest[batchsize:]
+        gpt_response_collection = gpt_response_collection[batchsize*2:]
+
+    ############################## <END> ##################################
+    history_to_return.extend(report_part_2)
+    res = write_results_to_file(history_to_return)
+    chatbot.append(("完成了吗？", res))
+    yield from update_ui(chatbot=chatbot, history=history_to_return) # 刷新界面
+
+
+@CatchException
+def 解析项目本身(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    history = []    # 清空历史，以免输入溢出
+    import glob
+    file_manifest = [f for f in glob.glob('./*.py') if ('test_project' not in f) and ('gpt_log' not in f)] + \
+                    [f for f in glob.glob('./crazy_functions/*.py') if ('test_project' not in f) and ('gpt_log' not in f)]+ \
+                    [f for f in glob.glob('./request_llm/*.py') if ('test_project' not in f) and ('gpt_log' not in f)]
+    project_folder = './'
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何python文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 解析源代码新(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)
+
+@CatchException
+def 解析一个Python项目(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.py', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何python文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 解析源代码新(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)
+
+
+@CatchException
+def 解析一个C项目的头文件(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.h', recursive=True)]  + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.hpp', recursive=True)] #+ \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.c', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.h头文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 解析源代码新(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)
+
+@CatchException
+def 解析一个C项目(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.h', recursive=True)]  + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.hpp', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.c', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.h头文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 解析源代码新(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)
+
+
+@CatchException
+def 解析一个Java项目(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    history = []  # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a=f"解析项目: {txt}", b=f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.java', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.jar', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.xml', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.sh', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a=f"解析项目: {txt}", b=f"找不到任何java文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 解析源代码新(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)
+
+
+@CatchException
+def 解析一个Rect项目(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    history = []  # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a=f"解析项目: {txt}", b=f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.ts', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.tsx', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.json', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.js', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.jsx', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a=f"解析项目: {txt}", b=f"找不到任何Rect文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 解析源代码新(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)
+
+
+@CatchException
+def 解析一个Golang项目(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    history = []  # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a=f"解析项目: {txt}", b=f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.go', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/go.mod', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/go.sum', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/go.work', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a=f"解析项目: {txt}", b=f"找不到任何golang文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 解析源代码新(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)
+    
+    
+@CatchException
+def 解析一个Lua项目(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.lua', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.xml', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.json', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.toml', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何lua文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 解析源代码新(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)    

crazy_functions/读文章写摘要.py

@@ -0,0 +1,67 @@
+from toolbox import update_ui
+from toolbox import CatchException, report_execption, write_results_to_file
+from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
+fast_debug = False
+
+
+def 解析Paper(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt):
+    import time, glob, os
+    print('begin analysis on:', file_manifest)
+    for index, fp in enumerate(file_manifest):
+        with open(fp, 'r', encoding='utf-8', errors='replace') as f:
+            file_content = f.read()
+
+        prefix = "接下来请你逐文件分析下面的论文文件，概括其内容" if index==0 else ""
+        i_say = prefix + f'请对下面的文章片段用中文做一个概述，文件名是{os.path.relpath(fp, project_folder)}，文章内容是 ```{file_content}```'
+        i_say_show_user = prefix + f'[{index}/{len(file_manifest)}] 请对下面的文章片段做一个概述: {os.path.abspath(fp)}'
+        chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+        if not fast_debug: 
+            msg = '正常'
+            # ** gpt request **
+            gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(i_say, i_say_show_user, llm_kwargs, chatbot, history=[], sys_prompt=system_prompt)   # 带超时倒计时
+
+            chatbot[-1] = (i_say_show_user, gpt_say)
+            history.append(i_say_show_user); history.append(gpt_say)
+            yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+            if not fast_debug: time.sleep(2)
+
+    all_file = ', '.join([os.path.relpath(fp, project_folder) for index, fp in enumerate(file_manifest)])
+    i_say = f'根据以上你自己的分析，对全文进行概括，用学术性语言写一段中文摘要，然后再写一段英文摘要（包括{all_file}）。'
+    chatbot.append((i_say, "[Local Message] waiting gpt response."))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    if not fast_debug: 
+        msg = '正常'
+        # ** gpt request **
+        gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(i_say, i_say, llm_kwargs, chatbot, history=history, sys_prompt=system_prompt)   # 带超时倒计时
+
+        chatbot[-1] = (i_say, gpt_say)
+        history.append(i_say); history.append(gpt_say)
+        yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+        res = write_results_to_file(history)
+        chatbot.append(("完成了吗？", res))
+        yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+
+
+
+@CatchException
+def 读文章写摘要(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)] # + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)] + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.c', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+    yield from 解析Paper(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)

crazy_functions/谷歌检索小助手.py

@@ -0,0 +1,106 @@
+from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
+from toolbox import CatchException, report_execption, write_results_to_file
+from toolbox import update_ui
+
+def get_meta_information(url, chatbot, history):
+    import requests
+    import arxiv
+    import difflib
+    from bs4 import BeautifulSoup
+    from toolbox import get_conf
+    proxies, = get_conf('proxies')
+    headers = {
+        'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/105.0.0.0 Safari/537.36',
+    }
+    # 发送 GET 请求
+    response = requests.get(url, proxies=proxies, headers=headers)
+
+    # 解析网页内容
+    soup = BeautifulSoup(response.text, "html.parser")
+
+    def string_similar(s1, s2):
+        return difflib.SequenceMatcher(None, s1, s2).quick_ratio()
+
+    profile = []
+    # 获取所有文章的标题和作者
+    for result in soup.select(".gs_ri"):
+        title = result.a.text.replace('\n', ' ').replace('  ', ' ')
+        author = result.select_one(".gs_a").text
+        try:
+            citation = result.select_one(".gs_fl > a[href*='cites']").text  # 引用次数是链接中的文本，直接取出来
+        except:
+            citation = 'cited by 0'
+        abstract = result.select_one(".gs_rs").text.strip()  # 摘要在 .gs_rs 中的文本，需要清除首尾空格
+        search = arxiv.Search(
+            query = title,
+            max_results = 1,
+            sort_by = arxiv.SortCriterion.Relevance,
+        )
+        paper = next(search.results())
+        if string_similar(title, paper.title) > 0.90: # same paper
+            abstract = paper.summary.replace('\n', ' ')
+            is_paper_in_arxiv = True
+        else:   # different paper
+            abstract = abstract
+            is_paper_in_arxiv = False
+        paper = next(search.results())
+        print(title)
+        print(author)
+        print(citation)
+        profile.append({
+            'title':title,
+            'author':author,
+            'citation':citation,
+            'abstract':abstract,
+            'is_paper_in_arxiv':is_paper_in_arxiv,
+        })
+
+        chatbot[-1] = [chatbot[-1][0], title + f'\n\n是否在arxiv中（不在arxiv中无法获取完整摘要）:{is_paper_in_arxiv}\n\n' + abstract]
+        yield from update_ui(chatbot=chatbot, history=[]) # 刷新界面
+    return profile
+
+@CatchException
+def 谷歌检索小助手(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    # 基本信息：功能、贡献者
+    chatbot.append([
+        "函数插件功能？",
+        "分析用户提供的谷歌学术（google scholar）搜索页面中，出现的所有文章: binary-husky，插件初始化中..."])
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import arxiv
+        from bs4 import BeautifulSoup
+    except:
+        report_execption(chatbot, history, 
+            a = f"解析项目: {txt}", 
+            b = f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade beautifulsoup4 arxiv```。")
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        return
+
+    # 清空历史，以免输入溢出
+    history = []
+
+    meta_paper_info_list = yield from get_meta_information(txt, chatbot, history)
+
+    if len(meta_paper_info_list[:10]) > 0:
+        i_say = "下面是一些学术文献的数据，请从中提取出以下内容。" + \
+        "1、英文题目；2、中文题目翻译；3、作者；4、arxiv公开（is_paper_in_arxiv）；4、引用数量（cite）；5、中文摘要翻译。" + \
+        f"以下是信息源：{str(meta_paper_info_list[:10])}" 
+
+        inputs_show_user = f"请分析此页面中出现的所有文章：{txt}"
+        gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
+            inputs=i_say, inputs_show_user=inputs_show_user, 
+            llm_kwargs=llm_kwargs, chatbot=chatbot, history=[], 
+            sys_prompt="你是一个学术翻译，请从数据中提取信息。你必须使用Markdown格式。你必须逐个文献进行处理。"
+        )
+
+        history.extend([ "第一批", gpt_say ])
+        meta_paper_info_list = meta_paper_info_list[10:]
+
+    chatbot.append(["状态？", "已经全部完成"])
+    msg = '正常'
+    yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面
+    res = write_results_to_file(history)
+    chatbot.append(("完成了吗？", res)); 
+    yield from update_ui(chatbot=chatbot, history=history, msg=msg) # 刷新界面

crazy_functions/高级功能函数模板.py

@@ -0,0 +1,29 @@
+from toolbox import CatchException, update_ui
+from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
+import datetime
+@CatchException
+def 高阶功能模板函数(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
+    """
+    txt             输入栏用户输入的文本，例如需要翻译的一段话，再例如一个包含了待处理文件的路径
+    llm_kwargs      gpt模型参数，如温度和top_p等，一般原样传递下去就行
+    plugin_kwargs   插件模型的参数，暂时没有用武之地
+    chatbot         聊天显示框的句柄，用于显示给用户
+    history         聊天历史，前情提要
+    system_prompt   给gpt的静默提醒
+    web_port        当前软件运行的端口号
+    """
+    history = []    # 清空历史，以免输入溢出
+    chatbot.append(("这是什么功能？", "[Local Message] 请注意，您正在调用一个[函数插件]的模板，该函数面向希望实现更多有趣功能的开发者，它可以作为创建新功能函数的模板（该函数只有20多行代码）。此外我们也提供可同步处理大量文件的多线程Demo供您参考。您若希望分享新的功能模组，请不吝PR！"))
+    yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # 由于请求gpt需要一段时间，我们先及时地做一次界面更新
+    for i in range(5):
+        currentMonth = (datetime.date.today() + datetime.timedelta(days=i)).month
+        currentDay = (datetime.date.today() + datetime.timedelta(days=i)).day
+        i_say = f'历史中哪些事件发生在{currentMonth}月{currentDay}日？列举两条并发送相关图片。发送图片时，请使用Markdown，将Unsplash API中的PUT_YOUR_QUERY_HERE替换成描述该事件的一个最重要的单词。'
+        gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
+            inputs=i_say, inputs_show_user=i_say, 
+            llm_kwargs=llm_kwargs, chatbot=chatbot, history=[], 
+            sys_prompt="当你想发送一张照片时，请使用Markdown, 并且不要有反斜线, 不要用代码块。使用 Unsplash API (https://source.unsplash.com/1280x720/? < PUT_YOUR_QUERY_HERE >)。"
+        )
+        chatbot[-1] = (i_say, gpt_say)
+        history.append(i_say);history.append(gpt_say)
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # 界面更新

functional.py

@@ -1,67 +0,0 @@
-# """
-# 'primary' for main call-to-action, 
-# 'secondary' for a more subdued style, 
-# 'stop' for a stop button.
-# """
-
-
-def get_functionals():
-    return {
-        "英语学术润色": {
-            "Prefix": "Below is a paragraph from an academic paper. Polish the writing to meet the academic style, \
-improve the spelling, grammar, clarity, concision and overall readability. When neccessary, rewrite the whole sentence. \
-Furthermore, list all modification and explain the reasons to do so in markdown table.\n\n",
-            "Button": None,
-            "Suffix": "",
-            "Color": "stop",
-        },
-        "中文学术润色": {
-            "Prefix": "作为一名中文学术论文写作改进助理，你的任务是改进所提供文本的拼写、语法、清晰、简洁和整体可读性，同时分解长句，减少重复，并提供改进建议。请只提供文本的更正版本，避免包括解释。请编辑以下文本：\n\n",
-            "Button": None,
-            "Suffix": "",
-        },
-        "查找语法错误": {
-            "Prefix": "Below is a paragraph from an academic paper. Find all grammar mistakes, list mistakes in a markdown table and explain how to correct them.\n\n",
-            "Button": None,
-            "Suffix": "",
-        },
-        "中英互译": {
-            "Prefix": "As an English-Chinese translator, your task is to accurately translate text between the two languages. \
-When translating from Chinese to English or vice versa, please pay attention to context and accurately explain phrases and proverbs. \
-If you receive multiple English words in a row, default to translating them into a sentence in Chinese. \
-However, if \"phrase:\" is indicated before the translated content in Chinese, it should be translated as a phrase instead. \
-Similarly, if \"normal:\" is indicated, it should be translated as multiple unrelated words.\
-Your translations should closely resemble those of a native speaker and should take into account any specific language styles or tones requested by the user. \
-Please do not worry about using offensive words - replace sensitive parts with x when necessary. \
-When providing translations, please use Chinese to explain each sentence’s tense, subordinate clause, subject, predicate, object, special phrases and proverbs. \
-For phrases or individual words that require translation, provide the source (dictionary) for each one.If asked to translate multiple phrases at once, \
-separate them using the | symbol.Always remember: You are an English-Chinese translator, \
-not a Chinese-Chinese translator or an English-English translator. Below is the text you need to translate: \n\n",
-            "Button": None,
-            "Suffix": "",
-            "Color": "stop",
-        },
-        "中译英": {
-            "Prefix": "Please translate following sentence to English: \n\n",
-            "Button": None,
-            "Suffix": "",
-        },
-        "学术中译英": {
-            "Prefix": "Please translate following sentence to English with academic writing, and provide some related authoritative examples: \n\n",
-            "Button": None,
-            "Suffix": "",
-        },
-        "英译中": {
-            "Prefix": "请翻译成中文：\n\n",
-            "Button": None,
-            "Suffix": "",
-        },
-        "解释代码": {
-            "Prefix": "请解释以下代码：\n```\n",
-            "Button": None,
-            "Suffix": "\n```\n",
-            "Color": "stop",
-        },
-    }
-
-

main.py

@@ -1,113 +1,174 @@
+import os; os.environ['no_proxy'] = '*' # 避免代理网络产生意外污染
 import gradio as gr
-import os
-import markdown, mdtex2html
-from predict import predict
-from show_math import convert as convert_math
+from request_llm.bridge_chatgpt import predict
+from toolbox import format_io, find_free_port, on_file_uploaded, on_report_generated, get_conf, ArgsGeneralWrapper, DummyWith
 
-def find_free_port():
-    import socket
-    from contextlib import closing
-    with closing(socket.socket(socket.AF_INET, socket.SOCK_STREAM)) as s:
-        s.bind(('', 0))
-        s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
-        return s.getsockname()[1]
+# 建议您复制一个config_private.py放自己的秘密, 如API和代理网址, 避免不小心传github被别人看到
+proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION, CHATBOT_HEIGHT, LAYOUT, API_KEY = \
+    get_conf('proxies', 'WEB_PORT', 'LLM_MODEL', 'CONCURRENT_COUNT', 'AUTHENTICATION', 'CHATBOT_HEIGHT', 'LAYOUT', 'API_KEY')
 
-PORT = find_free_port()
+# 如果WEB_PORT是-1, 则随机选取WEB端口
+PORT = find_free_port() if WEB_PORT <= 0 else WEB_PORT
+if not AUTHENTICATION: AUTHENTICATION = None
 
+from check_proxy import get_current_version
 initial_prompt = "Serve me as a writing and programming assistant."
-title_html = """<h1 align="center">ChatGPT 学术优化</h1>"""
+title_html = f"<h1 align=\"center\">ChatGPT 学术优化 {get_current_version()}</h1>"
+description =  """代码开源和更新[地址🚀](https://github.com/binary-husky/chatgpt_academic)，感谢热情的[开发者们❤️](https://github.com/binary-husky/chatgpt_academic/graphs/contributors)"""
 
+# 问询记录, python 版本建议3.9+（越新越好）
 import logging
-os.makedirs('gpt_log', exist_ok=True)
-logging.basicConfig(filename='gpt_log/predict.log', level=logging.INFO)
+os.makedirs("gpt_log", exist_ok=True)
+try:logging.basicConfig(filename="gpt_log/chat_secrets.log", level=logging.INFO, encoding="utf-8")
+except:logging.basicConfig(filename="gpt_log/chat_secrets.log", level=logging.INFO)
+print("所有问询记录将自动保存在本地目录./gpt_log/chat_secrets.log, 请注意自我隐私保护哦！")
 
+# 一些普通功能模块
+from core_functional import get_core_functions
+functional = get_core_functions()
 
-from functional import get_functionals
-functional = get_functionals()
-def reset_textbox(): return gr.update(value='')
+# 高级函数插件
+from crazy_functional import get_crazy_functions
+crazy_fns = get_crazy_functions()
 
-def text_divide_paragraph(text):
-    if '```' in text:
-        # careful input
-        return text
-    else:
-        # wtf input
-        lines = text.split("\n")
-        for i, line in enumerate(lines):
-            if i!=0: lines[i] = "<p>"+lines[i].replace(" ", "&nbsp;")+"</p>"
-        text = "".join(lines)
-        return text
-
-def markdown_convertion(txt):
-    if ('$' in txt) and ('```' not in txt):
-        math_config = {'mdx_math': {'enable_dollar_delimiter': True}}
-        return markdown.markdown(txt,extensions=['fenced_code','tables']) + '<br><br>' + \
-            markdown.markdown(convert_math(txt, splitParagraphs=False),extensions=['fenced_code','tables'])
-    else:
-        return markdown.markdown(txt,extensions=['fenced_code','tables'])
-
-# math_config = {'mdx_math': {'enable_dollar_delimiter': True}}
-# markdown.markdown(txt, extensions=['fenced_code', 'tables', 'mdx_math'], extension_configs=math_config)
-
-
-def format_io(self,y):
-    if y is None:
-        return []
-    i_ask, gpt_reply = y[-1]
-    
-    i_ask = text_divide_paragraph(i_ask) # 输入部分太自由，预处理一波
-    
-    y[-1] = (
-        None if i_ask is None else markdown.markdown(i_ask, extensions=['fenced_code','tables']),
-        None if gpt_reply is None else markdown_convertion(gpt_reply)
-    )
-    return y
+# 处理markdown文本格式的转变
 gr.Chatbot.postprocess = format_io
 
-with gr.Blocks() as demo:
+# 做一些外观色彩上的调整
+from theme import adjust_theme, advanced_css
+set_theme = adjust_theme()
+
+# 代理与自动更新
+from check_proxy import check_proxy, auto_update
+proxy_info = check_proxy(proxies)
+
+gr_L1 = lambda: gr.Row().style()
+gr_L2 = lambda scale: gr.Column(scale=scale)
+if LAYOUT == "TOP-DOWN": 
+    gr_L1 = lambda: DummyWith()
+    gr_L2 = lambda scale: gr.Row()
+    CHATBOT_HEIGHT /= 2
+
+cancel_handles = []
+with gr.Blocks(title="ChatGPT 学术优化", theme=set_theme, analytics_enabled=False, css=advanced_css) as demo:
     gr.HTML(title_html)
-    with gr.Row():
-        with gr.Column(scale=2):
+    cookies = gr.State({'api_key': API_KEY, 'llm_model': LLM_MODEL})
+    with gr_L1():
+        with gr_L2(scale=2):
             chatbot = gr.Chatbot()
-            chatbot.style(height=700)
-            chatbot.style()
+            chatbot.style(height=CHATBOT_HEIGHT)
             history = gr.State([])
-            TRUE = gr.State(True)
-            FALSE = gr.State(False)
-        with gr.Column(scale=1):
-            with gr.Row():
-                with gr.Column(scale=12):
+        with gr_L2(scale=1):
+            with gr.Accordion("输入区", open=True) as area_input_primary:
+                with gr.Row():
                     txt = gr.Textbox(show_label=False, placeholder="Input question here.").style(container=False)
-                with gr.Column(scale=1):
-                    submitBtn = gr.Button("Ask", variant="primary")
-            with gr.Row():
-                for k in functional:
-                    variant = functional[k]["Color"] if "Color" in functional[k] else "secondary"
-                    functional[k]["Button"] = gr.Button(k, variant=variant)
-
-            statusDisplay = gr.Markdown("status: ready")
-            systemPromptTxt = gr.Textbox(show_label=True, placeholder=f"System Prompt", label="System prompt", value=initial_prompt).style(container=True)
-            #inputs, top_p, temperature, top_k, repetition_penalty
-            with gr.Accordion("arguments", open=False):
+                with gr.Row():
+                    submitBtn = gr.Button("提交", variant="primary")
+                with gr.Row():
+                    resetBtn = gr.Button("重置", variant="secondary"); resetBtn.style(size="sm")
+                    stopBtn = gr.Button("停止", variant="secondary"); stopBtn.style(size="sm")
+                with gr.Row():
+                    status = gr.Markdown(f"Tip: 按Enter提交, 按Shift+Enter换行。当前模型: {LLM_MODEL} \n {proxy_info}")
+            with gr.Accordion("基础功能区", open=True) as area_basic_fn:
+                with gr.Row():
+                    for k in functional:
+                        variant = functional[k]["Color"] if "Color" in functional[k] else "secondary"
+                        functional[k]["Button"] = gr.Button(k, variant=variant)
+            with gr.Accordion("函数插件区", open=True) as area_crazy_fn:
+                with gr.Row():
+                    gr.Markdown("注意：以下“红颜色”标识的函数插件需从输入区读取路径作为参数.")
+                with gr.Row():
+                    for k in crazy_fns:
+                        if not crazy_fns[k].get("AsButton", True): continue
+                        variant = crazy_fns[k]["Color"] if "Color" in crazy_fns[k] else "secondary"
+                        crazy_fns[k]["Button"] = gr.Button(k, variant=variant)
+                        crazy_fns[k]["Button"].style(size="sm")
+                with gr.Row():
+                    with gr.Accordion("更多函数插件", open=True):
+                        dropdown_fn_list = [k for k in crazy_fns.keys() if not crazy_fns[k].get("AsButton", True)]
+                        with gr.Column(scale=1):
+                            dropdown = gr.Dropdown(dropdown_fn_list, value=r"打开插件列表", label="").style(container=False)
+                        with gr.Column(scale=1):
+                            switchy_bt = gr.Button(r"请先从插件列表中选择", variant="secondary")
+                with gr.Row():
+                    with gr.Accordion("点击展开“文件上传区”。上传本地文件可供红色函数插件调用。", open=False) as area_file_up:
+                        file_upload = gr.Files(label="任何文件, 但推荐上传压缩文件(zip, tar)", file_count="multiple")
+            with gr.Accordion("展开SysPrompt & 交互界面布局 & Github地址", open=(LAYOUT == "TOP-DOWN")):
+                system_prompt = gr.Textbox(show_label=True, placeholder=f"System Prompt", label="System prompt", value=initial_prompt)
                 top_p = gr.Slider(minimum=-0, maximum=1.0, value=1.0, step=0.01,interactive=True, label="Top-p (nucleus sampling)",)
-                temperature = gr.Slider(minimum=-0, maximum=5.0, value=1.0, step=0.01, interactive=True, label="Temperature",)
-
-    txt.submit(predict, [txt, top_p, temperature, chatbot, history, systemPromptTxt], [chatbot, history, statusDisplay])
-    submitBtn.click(predict, [txt, top_p, temperature, chatbot, history, systemPromptTxt], [chatbot, history, statusDisplay], show_progress=True)
-    # submitBtn.click(reset_textbox, [], [txt])
+                temperature = gr.Slider(minimum=-0, maximum=2.0, value=1.0, step=0.01, interactive=True, label="Temperature",)
+                checkboxes = gr.CheckboxGroup(["基础功能区", "函数插件区", "底部输入区"], value=["基础功能区", "函数插件区"], label="显示/隐藏功能区")
+                gr.Markdown(description)
+            with gr.Accordion("备选输入区", open=True, visible=False) as area_input_secondary:
+                with gr.Row():
+                    txt2 = gr.Textbox(show_label=False, placeholder="Input question here.", label="输入区2").style(container=False)
+                with gr.Row():
+                    submitBtn2 = gr.Button("提交", variant="primary")
+                with gr.Row():
+                    resetBtn2 = gr.Button("重置", variant="secondary"); resetBtn.style(size="sm")
+                    stopBtn2 = gr.Button("停止", variant="secondary"); stopBtn.style(size="sm")
+    # 功能区显示开关与功能区的互动
+    def fn_area_visibility(a):
+        ret = {}
+        ret.update({area_basic_fn: gr.update(visible=("基础功能区" in a))})
+        ret.update({area_crazy_fn: gr.update(visible=("函数插件区" in a))})
+        ret.update({area_input_primary: gr.update(visible=("底部输入区" not in a))})
+        ret.update({area_input_secondary: gr.update(visible=("底部输入区" in a))})
+        if "底部输入区" in a: ret.update({txt: gr.update(value="")})
+        return ret
+    checkboxes.select(fn_area_visibility, [checkboxes], [area_basic_fn, area_crazy_fn, area_input_primary, area_input_secondary, txt, txt2] )
+    # 整理反复出现的控件句柄组合
+    input_combo = [cookies, txt, txt2, top_p, temperature, chatbot, history, system_prompt]
+    output_combo = [cookies, chatbot, history, status]
+    predict_args = dict(fn=ArgsGeneralWrapper(predict), inputs=input_combo, outputs=output_combo)
+    # 提交按钮、重置按钮
+    cancel_handles.append(txt.submit(**predict_args))
+    cancel_handles.append(txt2.submit(**predict_args))
+    cancel_handles.append(submitBtn.click(**predict_args))
+    cancel_handles.append(submitBtn2.click(**predict_args))
+    resetBtn.click(lambda: ([], [], "已重置"), None, [chatbot, history, status])
+    resetBtn2.click(lambda: ([], [], "已重置"), None, [chatbot, history, status])
+    # 基础功能区的回调函数注册
     for k in functional:
-        functional[k]["Button"].click(predict, 
-            [txt, top_p, temperature, chatbot,history, systemPromptTxt, FALSE, TRUE, gr.State(k)], [chatbot, history, statusDisplay], show_progress=True)
-
-print(f"URL http://localhost:{PORT}")
-demo.title = "ChatGPT 学术优化"
-
+        click_handle = functional[k]["Button"].click(fn=ArgsGeneralWrapper(predict), inputs=[*input_combo, gr.State(True), gr.State(k)], outputs=output_combo)
+        cancel_handles.append(click_handle)
+    # 文件上传区，接收文件后与chatbot的互动
+    file_upload.upload(on_file_uploaded, [file_upload, chatbot, txt], [chatbot, txt])
+    # 函数插件-固定按钮区
+    for k in crazy_fns:
+        if not crazy_fns[k].get("AsButton", True): continue
+        click_handle = crazy_fns[k]["Button"].click(ArgsGeneralWrapper(crazy_fns[k]["Function"]), [*input_combo, gr.State(PORT)], output_combo)
+        click_handle.then(on_report_generated, [file_upload, chatbot], [file_upload, chatbot])
+        cancel_handles.append(click_handle)
+    # 函数插件-下拉菜单与随变按钮的互动
+    def on_dropdown_changed(k):
+        variant = crazy_fns[k]["Color"] if "Color" in crazy_fns[k] else "secondary"
+        return {switchy_bt: gr.update(value=k, variant=variant)}
+    dropdown.select(on_dropdown_changed, [dropdown], [switchy_bt] )
+    # 随变按钮的回调函数注册
+    def route(k, *args, **kwargs):
+        if k in [r"打开插件列表", r"请先从插件列表中选择"]: return 
+        yield from ArgsGeneralWrapper(crazy_fns[k]["Function"])(*args, **kwargs)
+    click_handle = switchy_bt.click(route,[switchy_bt, *input_combo, gr.State(PORT)], output_combo)
+    click_handle.then(on_report_generated, [file_upload, chatbot], [file_upload, chatbot])
+    # def expand_file_area(file_upload, area_file_up):
+    #     if len(file_upload)>0: return {area_file_up: gr.update(open=True)}
+    # click_handle.then(expand_file_area, [file_upload, area_file_up], [area_file_up])
+    cancel_handles.append(click_handle)
+    # 终止按钮的回调函数注册
+    stopBtn.click(fn=None, inputs=None, outputs=None, cancels=cancel_handles)
+    stopBtn2.click(fn=None, inputs=None, outputs=None, cancels=cancel_handles)
+# gradio的inbrowser触发不太稳定，回滚代码到原始的浏览器打开函数
 def auto_opentab_delay():
     import threading, webbrowser, time
-    def open(): time.sleep(2)
-    webbrowser.open_new_tab(f'http://localhost:{PORT}')
-    t = threading.Thread(target=open)
-    t.daemon = True; t.start()
+    print(f"如果浏览器没有自动打开，请复制并转到以下URL：")
+    print(f"\t（亮色主题）: http://localhost:{PORT}")
+    print(f"\t（暗色主题）: http://localhost:{PORT}/?__dark-theme=true")
+    def open(): 
+        time.sleep(2)       # 打开浏览器
+        webbrowser.open_new_tab(f"http://localhost:{PORT}/?__dark-theme=true")
+    threading.Thread(target=open, name="open-browser", daemon=True).start()
+    threading.Thread(target=auto_update, name="self-upgrade", daemon=True).start()
 
 auto_opentab_delay()
-demo.queue().launch(server_name="0.0.0.0", share=True, server_port=PORT)
+demo.queue(concurrency_count=CONCURRENT_COUNT).launch(server_name="0.0.0.0", server_port=PORT, auth=AUTHENTICATION)

predict.py

@@ -1,134 +0,0 @@
-import json
-import gradio as gr
-import logging
-import traceback
-import requests
-import importlib
-import os
-
-if os.path.exists('config_private.py'):
-    # 放自己的秘密如API和代理网址
-    from config_private import proxies, API_URL, API_KEY
-else:
-    from config import proxies, API_URL, API_KEY
-
-
-
-def compose_system(system_prompt):
-    return {"role": "system", "content": system_prompt}
-
-
-def compose_user(user_input):
-    return {"role": "user", "content": user_input}
-
-
-def predict(inputs, top_p, temperature, chatbot=[], history=[], system_prompt='', retry=False, 
-            stream = True, additional_fn=None):
-
-    if additional_fn is not None:
-        import functional
-        importlib.reload(functional)
-        functional = functional.get_functionals()
-        inputs = functional[additional_fn]["Prefix"] + inputs + functional[additional_fn]["Suffix"]
-
-    if stream:
-        raw_input = inputs
-        logging.info(f'[raw_input] {raw_input}')
-        chatbot.append((inputs, ""))
-        yield chatbot, history, "Waiting"
-
-    headers = {
-        "Content-Type": "application/json",
-        "Authorization": f"Bearer {API_KEY}"
-    }
-
-    chat_counter = len(history) // 2
-
-    print(f"chat_counter - {chat_counter}")
-
-    messages = [compose_system(system_prompt)]
-    if chat_counter:
-        for index in range(0, 2*chat_counter, 2):
-            d1 = {}
-            d1["role"] = "user"
-            d1["content"] = history[index]
-            d2 = {}
-            d2["role"] = "assistant"
-            d2["content"] = history[index+1]
-            if d1["content"] != "":
-                if d2["content"] != "" or retry:
-                    messages.append(d1)
-                    messages.append(d2)
-            else:
-                messages[-1]['content'] = d2['content']
-    if retry and chat_counter:
-        messages.pop()
-    else:
-        temp3 = {}
-        temp3["role"] = "user"
-        temp3["content"] = inputs
-        messages.append(temp3)
-        chat_counter += 1
-    # messages
-    payload = {
-        "model": "gpt-3.5-turbo",
-        # "model": "gpt-4",
-        "messages": messages, 
-        "temperature": temperature,  # 1.0,
-        "top_p": top_p,  # 1.0,
-        "n": 1,
-        "stream": stream,
-        "presence_penalty": 0,
-        "frequency_penalty": 0,
-    }
-
-    history.append(inputs)
-
-    try:
-        # make a POST request to the API endpoint using the requests.post method, passing in stream=True
-        response = requests.post(API_URL, headers=headers, proxies=proxies,
-                                json=payload, stream=True, timeout=15)
-    except:
-        chatbot[-1] = ((chatbot[-1][0], 'Requests Timeout, Network Error.'))
-        yield chatbot, history, "Requests Timeout"
-        raise TimeoutError
-
-    token_counter = 0
-    partial_words = ""
-
-    counter = 0
-    if stream:
-        stream_response =  response.iter_lines()
-        while True:
-            chunk = next(stream_response)
-            # print(chunk)
-            
-            if chunk == b'data: [DONE]':
-                break
-
-            if counter == 0:
-                counter += 1
-                continue
-            counter += 1
-            # check whether each line is non-empty
-            if chunk:
-                # decode each line as response data is in bytes
-                try:
-                    if len(json.loads(chunk.decode()[6:])['choices'][0]["delta"]) == 0:
-                        logging.info(f'[response] {chatbot[-1][-1]}')
-                        break
-                except Exception as e:
-                    traceback.print_exc()
-
-                chunkjson = json.loads(chunk.decode()[6:])
-                status_text = f"id: {chunkjson['id']}, finish_reason: {chunkjson['choices'][0]['finish_reason']}"
-                partial_words = partial_words + \
-                    json.loads(chunk.decode()[6:])[
-                        'choices'][0]["delta"]["content"]
-                if token_counter == 0:
-                    history.append(" " + partial_words)
-                else:
-                    history[-1] = partial_words
-                chatbot[-1] = (history[-2], history[-1])
-                token_counter += 1
-                yield chatbot, history, status_text

request_llm/README.md

@@ -0,0 +1,36 @@
+# 如何使用其他大语言模型（dev分支测试中）
+
+## 1. 先运行text-generation
+``` sh
+# 下载模型（ text-generation 这么牛的项目，别忘了给人家star ）
+git clone https://github.com/oobabooga/text-generation-webui.git
+
+# 安装text-generation的额外依赖
+pip install accelerate bitsandbytes flexgen gradio llamacpp markdown numpy peft requests rwkv safetensors sentencepiece tqdm datasets git+https://github.com/huggingface/transformers
+
+# 切换路径
+cd text-generation-webui
+
+# 下载模型
+python download-model.py facebook/galactica-1.3b
+# 其他可选如 facebook/opt-1.3b
+#           facebook/galactica-6.7b
+#           facebook/galactica-120b
+#           facebook/pygmalion-1.3b 等
+# 详情见 https://github.com/oobabooga/text-generation-webui
+
+# 启动text-generation，注意把模型的斜杠改成下划线
+python server.py --cpu --listen --listen-port 7860 --model facebook_galactica-1.3b
+```
+
+## 2. 修改config.py
+``` sh
+# LLM_MODEL格式较复杂   TGUI:[模型]@[ws地址]:[ws端口] ,   端口要和上面给定的端口一致
+LLM_MODEL = "TGUI:galactica-1.3b@localhost:7860"
+```
+
+## 3. 运行！
+``` sh
+cd chatgpt-academic
+python main.py
+```

request_llm/bridge_chatgpt.py

@@ -0,0 +1,253 @@
+# 借鉴了 https://github.com/GaiZhenbiao/ChuanhuChatGPT 项目
+
+"""
+    该文件中主要包含三个函数
+
+    不具备多线程能力的函数：
+    1. predict: 正常对话时使用，具备完备的交互功能，不可多线程
+
+    具备多线程调用能力的函数
+    2. predict_no_ui：高级实验性功能模块调用，不会实时显示在界面上，参数简单，可以多线程并行，方便实现复杂的功能逻辑
+    3. predict_no_ui_long_connection：在实验过程中发现调用predict_no_ui处理长文档时，和openai的连接容易断掉，这个函数用stream的方式解决这个问题，同样支持多线程
+"""
+
+import json
+import time
+import gradio as gr
+import logging
+import traceback
+import requests
+import importlib
+
+# config_private.py放自己的秘密如API和代理网址
+# 读取时首先看是否存在私密的config_private配置文件（不受git管控），如果有，则覆盖原config文件
+from toolbox import get_conf, update_ui
+proxies, API_URL, API_KEY, TIMEOUT_SECONDS, MAX_RETRY, LLM_MODEL = \
+    get_conf('proxies', 'API_URL', 'API_KEY', 'TIMEOUT_SECONDS', 'MAX_RETRY', 'LLM_MODEL')
+
+timeout_bot_msg = '[Local Message] Request timeout. Network error. Please check proxy settings in config.py.' + \
+                  '网络错误，检查代理服务器是否可用，以及代理设置的格式是否正确，格式须是[协议]://[地址]:[端口]，缺一不可。'
+
+def get_full_error(chunk, stream_response):
+    """
+        获取完整的从Openai返回的报错
+    """
+    while True:
+        try:
+            chunk += next(stream_response)
+        except:
+            break
+    return chunk
+
+
+def predict_no_ui_long_connection(inputs, llm_kwargs, history=[], sys_prompt="", observe_window=None, console_slience=False):
+    """
+        发送至chatGPT，等待回复，一次性完成，不显示中间过程。但内部用stream的方法避免中途网线被掐。
+        inputs：
+            是本次问询的输入
+        sys_prompt:
+            系统静默prompt
+        llm_kwargs：
+            chatGPT的内部调优参数
+        history：
+            是之前的对话列表
+        observe_window = None：
+            用于负责跨越线程传递已经输出的部分，大部分时候仅仅为了fancy的视觉效果，留空即可。observe_window[0]：观测窗。observe_window[1]：看门狗
+    """
+    watch_dog_patience = 5 # 看门狗的耐心, 设置5秒即可
+    headers, payload = generate_payload(inputs, llm_kwargs, history, system_prompt=sys_prompt, stream=True)
+    retry = 0
+    while True:
+        try:
+            # make a POST request to the API endpoint, stream=False
+            response = requests.post(API_URL, headers=headers, proxies=proxies,
+                                    json=payload, stream=True, timeout=TIMEOUT_SECONDS); break
+        except requests.exceptions.ReadTimeout as e:
+            retry += 1
+            traceback.print_exc()
+            if retry > MAX_RETRY: raise TimeoutError
+            if MAX_RETRY!=0: print(f'请求超时，正在重试 ({retry}/{MAX_RETRY}) ……')
+
+    stream_response =  response.iter_lines()
+    result = ''
+    while True:
+        try: chunk = next(stream_response).decode()
+        except StopIteration: 
+            break
+        except requests.exceptions.ConnectionError:
+            chunk = next(stream_response).decode() # 失败了，重试一次？再失败就没办法了。
+        if len(chunk)==0: continue
+        if not chunk.startswith('data:'): 
+            error_msg = get_full_error(chunk.encode('utf8'), stream_response).decode()
+            if "reduce the length" in error_msg:
+                raise ConnectionAbortedError("OpenAI拒绝了请求:" + error_msg)
+            else:
+                raise RuntimeError("OpenAI拒绝了请求：" + error_msg)
+        json_data = json.loads(chunk.lstrip('data:'))['choices'][0]
+        delta = json_data["delta"]
+        if len(delta) == 0: break
+        if "role" in delta: continue
+        if "content" in delta: 
+            result += delta["content"]
+            if not console_slience: print(delta["content"], end='')
+            if observe_window is not None: 
+                # 观测窗，把已经获取的数据显示出去
+                if len(observe_window) >= 1: observe_window[0] += delta["content"]
+                # 看门狗，如果超过期限没有喂狗，则终止
+                if len(observe_window) >= 2:  
+                    if (time.time()-observe_window[1]) > watch_dog_patience:
+                        raise RuntimeError("程序终止。")
+        else: raise RuntimeError("意外Json结构："+delta)
+    if json_data['finish_reason'] == 'length':
+        raise ConnectionAbortedError("正常结束，但显示Token不足，导致输出不完整，请削减单次输入的文本量。")
+    return result
+
+
+def predict(inputs, llm_kwargs, plugin_kwargs, chatbot, history=[], system_prompt='', stream = True, additional_fn=None):
+    """
+        发送至chatGPT，流式获取输出。
+        用于基础的对话功能。
+        inputs 是本次问询的输入
+        top_p, temperature是chatGPT的内部调优参数
+        history 是之前的对话列表（注意无论是inputs还是history，内容太长了都会触发token数量溢出的错误）
+        chatbot 为WebUI中显示的对话列表，修改它，然后yeild出去，可以直接修改对话界面内容
+        additional_fn代表点击的哪个按钮，按钮见functional.py
+    """
+    if inputs.startswith('sk-') and len(inputs) == 51:
+        chatbot._cookies['api_key'] = inputs
+        chatbot.append(("输入已识别为openai的api_key", "api_key已导入"))
+        yield from update_ui(chatbot=chatbot, history=history, msg="api_key已导入") # 刷新界面
+        return
+    elif len(chatbot._cookies['api_key']) != 51:
+        chatbot.append((inputs, "缺少api_key。\n\n1. 临时解决方案：直接在输入区键入api_key，然后回车提交。\n\n2. 长效解决方案：在config.py中配置。"))
+        yield from update_ui(chatbot=chatbot, history=history, msg="api_key已导入") # 刷新界面
+        return
+
+    if additional_fn is not None:
+        import core_functional
+        importlib.reload(core_functional)    # 热更新prompt
+        core_functional = core_functional.get_core_functions()
+        if "PreProcess" in core_functional[additional_fn]: inputs = core_functional[additional_fn]["PreProcess"](inputs)  # 获取预处理函数（如果有的话）
+        inputs = core_functional[additional_fn]["Prefix"] + inputs + core_functional[additional_fn]["Suffix"]
+
+    if stream:
+        raw_input = inputs
+        logging.info(f'[raw_input] {raw_input}')
+        chatbot.append((inputs, ""))
+        yield from update_ui(chatbot=chatbot, history=history, msg="等待响应") # 刷新界面
+
+    headers, payload = generate_payload(inputs, llm_kwargs, history, system_prompt, stream)
+    history.append(inputs); history.append(" ")
+
+    retry = 0
+    while True:
+        try:
+            # make a POST request to the API endpoint, stream=True
+            response = requests.post(API_URL, headers=headers, proxies=proxies,
+                                    json=payload, stream=True, timeout=TIMEOUT_SECONDS);break
+        except:
+            retry += 1
+            chatbot[-1] = ((chatbot[-1][0], timeout_bot_msg))
+            retry_msg = f"，正在重试 ({retry}/{MAX_RETRY}) ……" if MAX_RETRY > 0 else ""
+            yield from update_ui(chatbot=chatbot, history=history, msg="请求超时"+retry_msg) # 刷新界面
+            if retry > MAX_RETRY: raise TimeoutError
+
+    gpt_replying_buffer = ""
+    
+    is_head_of_the_stream = True
+    if stream:
+        stream_response =  response.iter_lines()
+        while True:
+            chunk = next(stream_response)
+            # print(chunk.decode()[6:])
+            if is_head_of_the_stream:
+                # 数据流的第一帧不携带content
+                is_head_of_the_stream = False; continue
+            
+            if chunk:
+                try:
+                    if len(json.loads(chunk.decode()[6:])['choices'][0]["delta"]) == 0:
+                        # 判定为数据流的结束，gpt_replying_buffer也写完了
+                        logging.info(f'[response] {gpt_replying_buffer}')
+                        break
+                    # 处理数据流的主体
+                    chunkjson = json.loads(chunk.decode()[6:])
+                    status_text = f"finish_reason: {chunkjson['choices'][0]['finish_reason']}"
+                    # 如果这里抛出异常，一般是文本过长，详情见get_full_error的输出
+                    gpt_replying_buffer = gpt_replying_buffer + json.loads(chunk.decode()[6:])['choices'][0]["delta"]["content"]
+                    history[-1] = gpt_replying_buffer
+                    chatbot[-1] = (history[-2], history[-1])
+                    yield from update_ui(chatbot=chatbot, history=history, msg=status_text) # 刷新界面
+
+                except Exception as e:
+                    traceback.print_exc()
+                    yield from update_ui(chatbot=chatbot, history=history, msg="Json解析不合常规") # 刷新界面
+                    chunk = get_full_error(chunk, stream_response)
+                    error_msg = chunk.decode()
+                    if "reduce the length" in error_msg:
+                        chatbot[-1] = (chatbot[-1][0], "[Local Message] Reduce the length. 本次输入过长，或历史数据过长. 历史缓存数据现已释放，您可以请再次尝试.")
+                        history = []    # 清除历史
+                    elif "Incorrect API key" in error_msg:
+                        chatbot[-1] = (chatbot[-1][0], "[Local Message] Incorrect API key. OpenAI以提供了不正确的API_KEY为由，拒绝服务.")
+                    elif "exceeded your current quota" in error_msg:
+                        chatbot[-1] = (chatbot[-1][0], "[Local Message] You exceeded your current quota. OpenAI以账户额度不足为由，拒绝服务.")
+                    else:
+                        from toolbox import regular_txt_to_markdown
+                        tb_str = '```\n' + traceback.format_exc() + '```'
+                        chatbot[-1] = (chatbot[-1][0], f"[Local Message] 异常 \n\n{tb_str} \n\n{regular_txt_to_markdown(chunk.decode()[4:])}")
+                    yield from update_ui(chatbot=chatbot, history=history, msg="Json异常" + error_msg) # 刷新界面
+                    return
+
+def generate_payload(inputs, llm_kwargs, history, system_prompt, stream):
+    """
+        整合所有信息，选择LLM模型，生成http请求，为发送请求做准备
+    """
+    if len(llm_kwargs['api_key']) != 51:
+        raise AssertionError("你提供了错误的API_KEY。\n\n1. 临时解决方案：直接在输入区键入api_key，然后回车提交。\n\n2. 长效解决方案：在config.py中配置。")
+
+    headers = {
+        "Content-Type": "application/json",
+        "Authorization": f"Bearer {llm_kwargs['api_key']}"
+    }
+
+    conversation_cnt = len(history) // 2
+
+    messages = [{"role": "system", "content": system_prompt}]
+    if conversation_cnt:
+        for index in range(0, 2*conversation_cnt, 2):
+            what_i_have_asked = {}
+            what_i_have_asked["role"] = "user"
+            what_i_have_asked["content"] = history[index]
+            what_gpt_answer = {}
+            what_gpt_answer["role"] = "assistant"
+            what_gpt_answer["content"] = history[index+1]
+            if what_i_have_asked["content"] != "":
+                if what_gpt_answer["content"] == "": continue
+                if what_gpt_answer["content"] == timeout_bot_msg: continue
+                messages.append(what_i_have_asked)
+                messages.append(what_gpt_answer)
+            else:
+                messages[-1]['content'] = what_gpt_answer['content']
+
+    what_i_ask_now = {}
+    what_i_ask_now["role"] = "user"
+    what_i_ask_now["content"] = inputs
+    messages.append(what_i_ask_now)
+
+    payload = {
+        "model": llm_kwargs['llm_model'],
+        "messages": messages, 
+        "temperature": llm_kwargs['temperature'],  # 1.0,
+        "top_p": llm_kwargs['top_p'],  # 1.0,
+        "n": 1,
+        "stream": stream,
+        "presence_penalty": 0,
+        "frequency_penalty": 0,
+    }
+    try:
+        print(f" {llm_kwargs['llm_model']} : {conversation_cnt} : {inputs[:100]} ..........")
+    except:
+        print('输入中可能存在乱码。')
+    return headers,payload
+
+

request_llm/bridge_tgui.py

@@ -0,0 +1,167 @@
+'''
+Contributed by SagsMug. Modified by binary-husky
+https://github.com/oobabooga/text-generation-webui/pull/175
+'''
+
+import asyncio
+import json
+import random
+import string
+import websockets
+import logging
+import time
+import threading
+import importlib
+from toolbox import get_conf, update_ui
+LLM_MODEL, = get_conf('LLM_MODEL')
+
+# "TGUI:galactica-1.3b@localhost:7860"
+model_name, addr_port = LLM_MODEL.split('@')
+assert ':' in addr_port, "LLM_MODEL 格式不正确！" + LLM_MODEL
+addr, port = addr_port.split(':')
+
+def random_hash():
+    letters = string.ascii_lowercase + string.digits
+    return ''.join(random.choice(letters) for i in range(9))
+
+async def run(context, max_token=512):
+    params = {
+        'max_new_tokens': max_token,
+        'do_sample': True,
+        'temperature': 0.5,
+        'top_p': 0.9,
+        'typical_p': 1,
+        'repetition_penalty': 1.05,
+        'encoder_repetition_penalty': 1.0,
+        'top_k': 0,
+        'min_length': 0,
+        'no_repeat_ngram_size': 0,
+        'num_beams': 1,
+        'penalty_alpha': 0,
+        'length_penalty': 1,
+        'early_stopping': True,
+        'seed': -1,
+    }
+    session = random_hash()
+
+    async with websockets.connect(f"ws://{addr}:{port}/queue/join") as websocket:
+        while content := json.loads(await websocket.recv()):
+            #Python3.10 syntax, replace with if elif on older
+            if content["msg"] ==  "send_hash":
+                await websocket.send(json.dumps({
+                    "session_hash": session,
+                    "fn_index": 12
+                }))
+            elif content["msg"] ==  "estimation":
+                pass
+            elif content["msg"] ==  "send_data":
+                await websocket.send(json.dumps({
+                    "session_hash": session,
+                    "fn_index": 12,
+                    "data": [
+                        context,
+                        params['max_new_tokens'],
+                        params['do_sample'],
+                        params['temperature'],
+                        params['top_p'],
+                        params['typical_p'],
+                        params['repetition_penalty'],
+                        params['encoder_repetition_penalty'],
+                        params['top_k'],
+                        params['min_length'],
+                        params['no_repeat_ngram_size'],
+                        params['num_beams'],
+                        params['penalty_alpha'],
+                        params['length_penalty'],
+                        params['early_stopping'],
+                        params['seed'],
+                    ]
+                }))
+            elif content["msg"] ==  "process_starts":
+                pass
+            elif content["msg"] in ["process_generating", "process_completed"]:
+                yield content["output"]["data"][0]
+                # You can search for your desired end indicator and 
+                #  stop generation by closing the websocket here
+                if (content["msg"] == "process_completed"):
+                    break
+
+
+
+
+
+def predict_tgui(inputs, top_p, temperature, chatbot, history=[], system_prompt='', stream = True, additional_fn=None):
+    """
+        发送至chatGPT，流式获取输出。
+        用于基础的对话功能。
+        inputs 是本次问询的输入
+        top_p, temperature是chatGPT的内部调优参数
+        history 是之前的对话列表（注意无论是inputs还是history，内容太长了都会触发token数量溢出的错误）
+        chatbot 为WebUI中显示的对话列表，修改它，然后yeild出去，可以直接修改对话界面内容
+        additional_fn代表点击的哪个按钮，按钮见functional.py
+    """
+    if additional_fn is not None:
+        import core_functional
+        importlib.reload(core_functional)    # 热更新prompt
+        core_functional = core_functional.get_core_functions()
+        if "PreProcess" in core_functional[additional_fn]: inputs = core_functional[additional_fn]["PreProcess"](inputs)  # 获取预处理函数（如果有的话）
+        inputs = core_functional[additional_fn]["Prefix"] + inputs + core_functional[additional_fn]["Suffix"]
+
+    raw_input = "What I would like to say is the following: " + inputs
+    logging.info(f'[raw_input] {raw_input}')
+    history.extend([inputs, ""])
+    chatbot.append([inputs, ""])
+    yield from update_ui(chatbot=chatbot, history=history, msg="等待响应") # 刷新界面
+
+    prompt = inputs
+    tgui_say = ""
+
+    mutable = ["", time.time()]
+    def run_coorotine(mutable):
+        async def get_result(mutable):
+            async for response in run(prompt):
+                print(response[len(mutable[0]):])
+                mutable[0] = response
+                if (time.time() - mutable[1]) > 3: 
+                    print('exit when no listener')
+                    break
+        asyncio.run(get_result(mutable))
+
+    thread_listen = threading.Thread(target=run_coorotine, args=(mutable,), daemon=True)
+    thread_listen.start()
+
+    while thread_listen.is_alive():
+        time.sleep(1)
+        mutable[1] = time.time()
+        # Print intermediate steps
+        if tgui_say != mutable[0]:
+            tgui_say = mutable[0]
+            history[-1] = tgui_say
+            chatbot[-1] = (history[-2], history[-1])
+            yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+
+    logging.info(f'[response] {tgui_say}')
+
+
+
+def predict_tgui_no_ui(inputs, top_p, temperature, history=[], sys_prompt=""):
+    raw_input = "What I would like to say is the following: " + inputs
+    prompt = inputs
+    tgui_say = ""
+    mutable = ["", time.time()]
+    def run_coorotine(mutable):
+        async def get_result(mutable):
+            async for response in run(prompt, max_token=20):
+                print(response[len(mutable[0]):])
+                mutable[0] = response
+                if (time.time() - mutable[1]) > 3: 
+                    print('exit when no listener')
+                    break
+        asyncio.run(get_result(mutable))
+    thread_listen = threading.Thread(target=run_coorotine, args=(mutable,))
+    thread_listen.start()
+    while thread_listen.is_alive():
+        time.sleep(1)
+        mutable[1] = time.time()
+    tgui_say = mutable[0]
+    return tgui_say

requirements.txt

@@ -1,3 +1,16 @@
-gradio
+gradio>=3.25.0
+tiktoken>=0.3.3
 requests[socks]
+transformers
+python-markdown-math
+beautifulsoup4
+latex2mathml
+python-docx
 mdtex2html
+colorama
+Markdown
+pygments
+pymupdf
+openai
+numpy
+arxiv

self_analysis.md

@@ -0,0 +1,262 @@
+# chatgpt-academic项目自译解报告
+（Author补充：以下分析均由本项目调用ChatGPT一键生成，如果有不准确的地方，全怪GPT😄）
+
+## 对程序的整体功能和构架做出概括。然后用一张markdown表格整理每个文件的功能（包括'check_proxy.py', 'config.py'等）。
+
+整体概括：
+
+该程序是一个基于自然语言处理和机器学习的科学论文辅助工具，主要功能包括聊天机器人、批量总结PDF文档、批量翻译PDF文档、生成函数注释、解析项目源代码等。程序基于 Gradio 构建 Web 服务，并集成了代理和自动更新功能，提高了用户的使用体验。
+
+文件功能表格：
+
+| 文件名称                                                     | 功能                                                         |
+| ------------------------------------------------------------ | ------------------------------------------------------------ |
+| .\check_proxy.py                                             | 检查代理设置功能。                                           |
+| .\config.py                                                  | 配置文件，存储程序的基本设置。                               |
+| .\config_private.py                                          | 存储代理网络地址的文件。                                     |
+| .\core_functional.py                                         | 主要的程序逻辑，包括聊天机器人和文件处理。                  |
+| .\cradle.py                                                  | 程序入口，初始化程序和启动 Web 服务。                        |
+| .\crazy_functional.py                                        | 辅助程序功能，包括PDF文档处理、代码处理、函数注释生成等。  |
+| .\main.py                                                    | 包含聊天机器人的具体实现。                                   |
+| .\show_math.py                                               | 处理 LaTeX 公式的函数。                                      |
+| .\theme.py                                                   | 存储 Gradio Web 服务的 CSS 样式文件。                        |
+| .\toolbox.py                                                 | 提供了一系列工具函数，包括文件读写、网页抓取、解析函数参数、生成 HTML 等。 |
+| ./crazy_functions/crazy_utils.py                             | 提供各种工具函数，如解析字符串、清洗文本、清理目录结构等。 |
+| ./crazy_functions/\_\_init\_\_.py                             | crazy_functions 模块的入口文件。                            |
+| ./crazy_functions/下载arxiv论文翻译摘要.py                 | 对 arxiv.org 上的 PDF 论文进行下载和翻译。                   |
+| ./crazy_functions/代码重写为全英文_多线程.py                | 将代码文件中的中文注释和字符串替换为英文。                   |
+| ./crazy_functions/总结word文档.py                            | 读取 Word 文档并生成摘要。                                    |
+| ./crazy_functions/批量总结PDF文档.py                        | 批量读取 PDF 文件并生成摘要。                                 |
+| ./crazy_functions/批量总结PDF文档pdfminer.py                | 使用 pdfminer 库进行 PDF 文件处理。                        |
+| ./crazy_functions/批量翻译PDF文档_多线程.py                 | 使用多线程技术批量翻译 PDF 文件。                             |
+| ./crazy_functions/生成函数注释.py                           | 给 Python 函数自动生成说明文档。                             |
+| ./crazy_functions/解析项目源代码.py                         | 解析项目中的源代码，提取注释和函数名等信息。                  |
+| ./crazy_functions/读文章写摘要.py                           | 读取多个文本文件并生成对应的摘要。                             |
+| ./crazy_functions/高级功能函数模板.py                        | 使用 GPT 模型进行文本处理。                                  |
+
+
+
+## [0/22] 程序概述: check_proxy.py
+
+该程序的文件名是check_proxy.py，主要有两个函数：check_proxy和auto_update。
+
+check_proxy函数中会借助requests库向一个IP查询API发送请求，并返回该IP的地理位置信息。同时根据返回的数据来判断代理是否有效。
+
+auto_update函数主要用于检查程序更新，会从Github获取程序最新的版本信息，如果当前版本和最新版本相差较大，则会提示用户进行更新。该函数中也会依赖requests库进行网络请求。
+
+在程序的开头，还添加了一句防止代理网络影响的代码。程序使用了自己编写的toolbox模块中的get_conf函数来获取代理设置。
+
+## [1/22] 程序概述: config.py
+
+该程序文件是一个Python模块，文件名为config.py。该模块包含了一些变量和配置选项，用于配置一个OpenAI的聊天机器人。具体的配置选项如下：
+
+- API_KEY: 密钥，用于连接OpenAI的API。需要填写有效的API密钥。
+- USE_PROXY: 是否使用代理。如果需要使用代理，需要将其改为True。
+- proxies: 代理的协议、地址和端口。
+- CHATBOT_HEIGHT: 聊天机器人对话框的高度。
+- LAYOUT: 聊天机器人对话框的布局，默认为左右布局。
+- TIMEOUT_SECONDS: 发送请求到OpenAI后，等待多久判定为超时。
+- WEB_PORT: 网页的端口，-1代表随机端口。
+- MAX_RETRY: 如果OpenAI不响应（网络卡顿、代理失败、KEY失效），重试的次数限制。
+- LLM_MODEL: OpenAI模型选择，目前只对某些用户开放的gpt4。
+- API_URL: OpenAI的API地址。
+- CONCURRENT_COUNT: 使用的线程数。
+- AUTHENTICATION: 用户名和密码，如果需要。
+
+## [2/22] 程序概述: config_private.py
+
+该程序文件名为config_private.py，包含了API_KEY的设置和代理的配置。使用了一个名为API_KEY的常量来存储私人的API密钥。此外，还有一个名为USE_PROXY的常量来标记是否需要使用代理。如果需要代理，则使用了一个名为proxies的字典来存储代理网络的地址，其中包括协议类型、地址和端口。
+
+## [3/22] 程序概述: core_functional.py
+
+该程序文件名为`core_functional.py`，主要是定义了一些核心功能函数，包括英语和中文学术润色、查找语法错误、中译英、学术中英互译、英译中、找图片和解释代码等。每个功能都有一个`Prefix`属性和`Suffix`属性，`Prefix`是指在用户输入的任务前面要显示的文本，`Suffix`是指在任务后面要显示的文本。此外，还有一个`Color`属性指示按钮的颜色，以及一个`PreProcess`函数表示对输入进行预处理的函数。
+
+## [4/22] 程序概述: cradle.py
+
+该程序文件名为cradle.py，主要功能是检测当前版本与远程最新版本是否一致，如果不一致则输出新版本信息并提示更新。其流程大致如下：
+
+1. 导入相关模块与自定义工具箱函数get_conf 
+2. 读取配置文件中的代理proxies 
+3. 使用requests模块请求远程版本信息（url为https://raw.githubusercontent.com/binary-husky/chatgpt_academic/master/version）并加载为json格式 
+4. 获取远程版本号、是否显示新功能信息、新功能内容 
+5. 读取本地版本文件version并加载为json格式 
+6. 获取当前版本号 
+7. 比较当前版本与远程版本，如果远程版本号比当前版本号高0.05以上，则输出新版本信息并提示更新 
+8. 如果不需要更新，则直接返回
+
+## [5/22] 程序概述: crazy_functional.py
+
+该程序文件名为.\crazy_functional.py，主要定义了一个名为get_crazy_functions()的函数，该函数返回一个字典类型的变量function_plugins，其中包含了一些函数插件。
+
+一些重要的函数插件包括：
+
+- 读文章写摘要：可以自动读取Tex格式的论文，并生成其摘要。
+
+- 批量生成函数注释：可以批量生成Python函数的文档注释。
+
+- 解析项目源代码：可以解析Python、C++、Golang、Java及React项目的源代码。
+
+- 批量总结PDF文档：可以对PDF文档进行批量总结，以提取其中的关键信息。
+
+- 一键下载arxiv论文并翻译摘要：可以自动下载arxiv.org网站上的PDF论文，并翻译生成其摘要。
+
+- 批量翻译PDF文档（多线程）：可以对PDF文档进行批量翻译，并使用多线程方式提高翻译效率。
+
+## [6/22] 程序概述: main.py
+
+本程序为一个基于 Gradio 和 GPT-3 的交互式聊天机器人，文件名为 main.py。其中主要功能包括：
+
+1. 使用 Gradio 建立 Web 界面，实现用户与聊天机器人的交互；
+2. 通过 bridge_chatgpt 模块，利用 GPT-3 模型实现聊天机器人的逻辑；
+3. 提供一些基础功能和高级函数插件，用户可以通过按钮选择使用；
+4. 提供文档格式转变、外观调整以及代理和自动更新等功能。
+
+程序的主要流程为：
+
+1. 导入所需的库和模块，并通过 get_conf 函数获取配置信息；
+2. 设置 Gradio 界面的各个组件，包括聊天窗口、输入区、功能区、函数插件区等；
+3. 注册各个组件的回调函数，包括用户输入、信号按钮等，实现机器人逻辑的交互；
+4. 通过 Gradio 的 queue 函数和 launch 函数启动 Web 服务，并提供聊天机器人的功能。
+
+此外，程序还提供了代理和自动更新功能，可以确保用户的使用体验。
+
+## [7/22] 程序概述: show_math.py
+
+该程序是一个Python脚本，文件名为show_math.py。它转换Markdown和LaTeX混合语法到带MathML的HTML。程序使用latex2mathml模块来实现从LaTeX到MathML的转换，将符号转换为HTML实体以批量处理。程序利用正则表达式和递归函数的方法处理不同形式的LaTeX语法，支持以下四种情况：$$形式、$形式、\[..]形式和\(...\)形式。如果无法转换某个公式，则在该位置插入一条错误消息。最后，程序输出HTML字符串。
+
+## [8/22] 程序概述: theme.py
+
+该程序文件为一个Python脚本，其功能是调整Gradio应用的主题和样式，包括字体、颜色、阴影、背景等等。在程序中，使用了Gradio提供的默认颜色主题，并针对不同元素设置了相应的样式属性，以达到美化显示的效果。此外，程序中还包含了一段高级CSS样式代码，针对表格、列表、聊天气泡、行内代码等元素进行了样式设定。
+
+## [9/22] 程序概述: toolbox.py
+
+此程序文件主要包含了一系列用于聊天机器人开发的实用工具函数和装饰器函数。主要函数包括：
+
+1. ArgsGeneralWrapper：一个装饰器函数，用于重组输入参数，改变输入参数的顺序与结构。
+
+2. get_reduce_token_percent：一个函数，用于计算自然语言处理时会出现的token溢出比例。
+
+3. predict_no_ui_but_counting_down：一个函数，调用聊天接口，并且保留了一定的界面心跳功能，即当对话太长时，会自动采用二分法截断。
+
+4. write_results_to_file：一个函数，将对话记录history生成Markdown格式的文本，并写入文件中。
+
+5. regular_txt_to_markdown：一个函数，将普通文本转换为Markdown格式的文本。
+
+6. CatchException：一个装饰器函数，捕捉函数调度中的异常，并封装到一个生成器中返回，并显示到聊天当中。
+
+7. HotReload：一个装饰器函数，实现函数插件的热更新。
+
+8. report_execption：一个函数，向chatbot中添加错误信息。
+
+9. text_divide_paragraph：一个函数，将文本按照段落分隔符分割开，生成带有段落标签的HTML代码。
+
+10. markdown_convertion：一个函数，将Markdown格式的文本转换为HTML格式。如果包含数学公式，则先将公式转换为HTML格式。
+
+11. close_up_code_segment_during_stream：一个函数，用于在gpt输出代码的中途，即输出了前面的```，但还没输出完后面的```，补上后面的```。
+
+12. format_io：一个函数，将输入和输出解析为HTML格式。将输出部分的Markdown和数学公式转换为HTML格式。
+
+13. find_free_port：一个函数，返回当前系统中可用的未使用端口。
+
+14. extract_archive：一个函数，解压缩文件。
+
+15. find_recent_files：一个函数，查找目录下一分钟内创建的文件。
+
+16. on_file_uploaded：一个函数，响应用户上传的文件。
+
+## [10/22] 程序概述: crazy_functions\crazy_utils.py
+
+这是一个名为"crazy_utils.py"的Python程序文件，包含了两个函数：
+1. `breakdown_txt_to_satisfy_token_limit()`：接受文本字符串、计算文本单词数量的函数和单词数量限制作为输入参数，将长文本拆分成合适的长度，以满足单词数量限制。这个函数使用一个递归方法去拆分长文本。
+2. `breakdown_txt_to_satisfy_token_limit_for_pdf()`：类似于`breakdown_txt_to_satisfy_token_limit()`，但是它使用一个不同的递归方法来拆分长文本，以满足PDF文档中的需求。当出现无法继续拆分的情况时，该函数将使用一个中文句号标记插入文本来截断长文本。如果还是无法拆分，则会引发运行时异常。
+
+## [11/22] 程序概述: crazy_functions\__init__.py
+
+这个程序文件是一个 Python 的包，包名为 "crazy_functions"，并且是其中的一个子模块 "__init__.py"。该包中可能包含多个函数或类，用于实现各种疯狂的功能。由于该文件的具体代码没有给出，因此无法进一步确定该包中的功能。通常情况下，一个包应该具有 __init__.py、__main__.py 和其它相关的模块文件，用于实现该包的各种功能。
+
+## [12/22] 程序概述: crazy_functions\下载arxiv论文翻译摘要.py
+
+这个程序实现的功能是下载arxiv论文并翻译摘要，文件名为`下载arxiv论文翻译摘要.py`。这个程序引入了`requests`、`unicodedata`、`os`、`re`等Python标准库，以及`pdfminer`、`bs4`等第三方库。其中`download_arxiv_`函数主要实现了从arxiv网站下载论文的功能，包括解析链接、获取论文信息、下载论文和生成文件名等，`get_name`函数则是为了从arxiv网站中获取论文信息创建的辅助函数。`下载arxiv论文并翻译摘要`函数则是实现了从下载好的PDF文件中提取摘要，然后使用预先训练的GPT模型翻译为中文的功能。同时，该函数还会将历史记录写入文件中。函数还会通过`CatchException`函数来捕获程序中出现的异常信息。
+
+## [13/22] 程序概述: crazy_functions\代码重写为全英文_多线程.py
+
+该程序文件为一个Python多线程程序，文件名为"crazy_functions\代码重写为全英文_多线程.py"。该程序使用了多线程技术，将一个大任务拆成多个小任务，同时执行，提高运行效率。
+
+程序的主要功能是将Python文件中的中文转换为英文，同时将转换后的代码输出。程序先清空历史记录，然后尝试导入openai和transformers等依赖库。程序接下来会读取当前路径下的.py文件和crazy_functions文件夹中的.py文件，并将其整合成一个文件清单。随后程序会使用GPT2模型进行中英文的翻译，并将结果保存在本地路径下的"gpt_log/generated_english_version"文件夹中。程序最终会生成一个任务执行报告。
+
+需要注意的是，该程序依赖于"request_llm"和"toolbox"库以及本地的"crazy_utils"模块。
+
+## [14/22] 程序概述: crazy_functions\总结word文档.py
+
+该程序文件是一个 Python 脚本文件，文件名为 ./crazy_functions/总结word文档.py。该脚本是一个函数插件，提供了名为“总结word文档”的函数。该函数的主要功能是批量读取给定文件夹下的 Word 文档文件，并使用 GPT 模型生成对每个文件的概述和意见建议。其中涉及到了读取 Word 文档、使用 GPT 模型等操作，依赖于许多第三方库。该文件也提供了导入依赖的方法，使用该脚本需要安装依赖库 python-docx 和 pywin32。函数功能实现的过程中，使用了一些用于调试的变量（如 fast_debug），可在需要时设置为 True。该脚本文件也提供了对程序功能和贡献者的注释。
+
+## [15/22] 程序概述: crazy_functions\批量总结PDF文档.py
+
+该程序文件名为 `./crazy_functions\批量总结PDF文档.py`，主要实现了批量处理PDF文档的功能。具体实现了以下几个函数：
+
+1. `is_paragraph_break(match)`：根据给定的匹配结果判断换行符是否表示段落分隔。
+2. `normalize_text(text)`：通过将文本特殊符号转换为其基本形式来对文本进行归一化处理。
+3. `clean_text(raw_text)`：对从 PDF 提取出的原始文本进行清洗和格式化处理。
+4. `解析PDF(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)`：对给定的PDF文件进行分析并生成相应的概述。
+5. `批量总结PDF文档(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT)`：批量处理PDF文件，对其进行摘要生成。
+
+其中，主要用到了第三方库`pymupdf`对PDF文件进行处理。程序通过调用`fitz.open`函数打开PDF文件，使用`page.get_text()`方法获取PDF文本内容。然后，使用`clean_text`函数对文本进行清洗和格式化处理，生成最终的摘要。最后，调用`write_results_to_file`函数将历史记录写入文件并输出。
+
+## [16/22] 程序概述: crazy_functions\批量总结PDF文档pdfminer.py
+
+这个程序文件名是./crazy_functions\批量总结PDF文档pdfminer.py，是一个用于批量读取PDF文件，解析其中的内容，并对其进行概括的程序。程序中引用了pdfminer和beautifulsoup4等Python库，读取PDF文件并将其转化为文本内容，然后利用GPT模型生成摘要语言，最终输出一个中文和英文的摘要。程序还有一些错误处理的代码，会输出错误信息。
+
+## [17/22] 程序概述: crazy_functions\批量翻译PDF文档_多线程.py
+
+这是一个 Python 程序文件，文件名为 `批量翻译PDF文档_多线程.py`，包含多个函数。主要功能是批量处理 PDF 文档，解析其中的文本，进行清洗和格式化处理，并使用 OpenAI 的 GPT 模型进行翻译。其中使用了多线程技术来提高程序的效率和并行度。
+
+## [18/22] 程序概述: crazy_functions\生成函数注释.py
+
+该程序文件名为./crazy_functions\生成函数注释.py。该文件包含两个函数，分别为`生成函数注释`和`批量生成函数注释`。
+
+函数`生成函数注释`包含参数`file_manifest`、`project_folder`、`top_p`、`temperature`、`chatbot`、`history`和`systemPromptTxt`。其中，`file_manifest`为一个包含待处理文件路径的列表，`project_folder`表示项目文件夹路径，`top_p`和`temperature`是GPT模型参数，`chatbot`为与用户交互的聊天机器人，`history`记录聊天机器人与用户的历史记录，`systemPromptTxt`为聊天机器人发送信息前的提示语。`生成函数注释`通过读取文件内容，并调用GPT模型对文件中的所有函数生成注释，最后使用markdown表格输出结果。函数中还包含一些条件判断和计时器，以及调用其他自定义模块的函数。
+
+函数`批量生成函数注释`包含参数`txt`、`top_p`、`temperature`、`chatbot`、`history`、`systemPromptTxt`和`WEB_PORT`。其中，`txt`表示用户输入的项目文件夹路径，其他参数含义与`生成函数注释`中相同。`批量生成函数注释`主要是通过解析项目文件夹，获取所有待处理文件的路径，并调用函数`生成函数注释`对每个文件进行处理，最终生成注释表格输出给用户。
+
+## [19/22] 程序概述: crazy_functions\解析项目源代码.py
+
+该程序文件包含了多个函数，用于解析不同类型的项目，如Python项目、C项目、Java项目等。其中，最核心的函数是`解析源代码()`，它会对给定的一组文件进行分析，并返回对应的结果。具体流程如下：
+
+1. 遍历所有待分析的文件，对每个文件进行如下处理：
+
+   1.1 从文件中读取代码内容，构造成一个字符串。
+
+   1.2 构造一条GPT请求，向`predict_no_ui_but_counting_down()`函数发送请求，等待GPT回复。
+
+   1.3 将GPT回复添加到机器人会话列表中，更新历史记录。
+
+   1.4 如果不是快速调试模式，则等待2秒钟，继续分析下一个文件。
+
+2. 如果所有文件都分析完成，则向机器人会话列表中添加一条新消息，提示用户整个分析过程已经结束。
+
+3. 返回机器人会话列表和历史记录。
+
+除此之外，该程序文件还定义了若干个函数，用于针对不同类型的项目进行解析。这些函数会按照不同的方式调用`解析源代码()`函数。例如，对于Python项目，只需要分析.py文件；对于C项目，需要同时分析.h和.cpp文件等。每个函数中都会首先根据给定的项目路径读取相应的文件，然后调用`解析源代码()`函数进行分析。
+
+## [20/22] 程序概述: crazy_functions\读文章写摘要.py
+
+该程序文件为一个名为“读文章写摘要”的Python函数，用于解析项目文件夹中的.tex文件，并使用GPT模型生成文章的中英文摘要。函数使用了request_llm.bridge_chatgpt和toolbox模块中的函数，并包含两个子函数：解析Paper和CatchException。函数参数包括txt，top_p，temperature，chatbot，history，systemPromptTxt和WEB_PORT。执行过程中函数首先清空历史，然后根据项目文件夹中的.tex文件列表，对每个文件调用解析Paper函数生成中文摘要，最后根据所有文件的中文摘要，调用GPT模型生成英文摘要。函数运行过程中会将结果写入文件并返回聊天机器人和历史记录。
+
+## [21/22] 程序概述: crazy_functions\高级功能函数模板.py
+
+该程序文件为一个高级功能函数模板，文件名为"./crazy_functions\高级功能函数模板.py"。
+
+该文件导入了两个模块，分别是"request_llm.bridge_chatgpt"和"toolbox"。其中"request_llm.bridge_chatgpt"模块包含了一个函数"predict_no_ui_long_connection"，该函数用于请求GPT模型进行对话生成。"toolbox"模块包含了三个函数，分别是"catchException"、"report_exception"和"write_results_to_file"函数，这三个函数主要用于异常处理和日志记录等。
+
+该文件定义了一个名为"高阶功能模板函数"的函数，并通过"decorator"装饰器将该函数装饰为一个异常处理函数，可以处理函数执行过程中出现的错误。该函数的作用是生成历史事件查询的问题，并向用户询问历史中哪些事件发生在指定日期，并索要相关图片。在查询完所有日期后，该函数返回所有历史事件及其相关图片的列表。其中，该函数的输入参数包括：
+
+1. txt: 一个字符串，表示当前消息的文本内容。
+2. top_p: 一个浮点数，表示GPT模型生成文本时的"top_p"参数。
+3. temperature: 一个浮点数，表示GPT模型生成文本时的"temperature"参数。
+4. chatbot: 一个列表，表示当前对话的记录列表。
+5. history: 一个列表，表示当前对话的历史记录列表。
+6. systemPromptTxt: 一个字符串，表示当前对话的系统提示信息。
+7. WEB_PORT: 一个整数，表示当前应用程序的WEB端口号。
+
+该函数在执行过程中，会先清空历史记录，以免输入溢出。然后，它会循环5次，生成5个历史事件查询的问题，并向用户请求输入相关信息。每次询问不携带之前的询问历史。在生成每个问题时，该函数会向"chatbot"列表中添加一条消息记录，并设置该记录的初始状态为"[Local Message] waiting gpt response."。然后，该函数会调用"predict_no_ui_long_connection"函数向GPT模型请求生成一段文本，并将生成的文本作为回答。如果请求过程中出现异常，该函数会忽略异常。最后，该函数将问题和回答添加到"chatbot"列表和"history"列表中，并将"chatbot"和"history"列表作为函数的返回值返回。
+

show_math.py

@@ -1,78 +0,0 @@
-from latex2mathml.converter import convert as tex2mathml
-import re
-
-incomplete = '<font style="color:orange;" class="tooltip">&#9888;<span class="tooltiptext">formula incomplete</span></font>'
-convError = '<font style="color:red" class="tooltip">&#9888;<span class="tooltiptext">LaTeX-convert-error</span></font>'
-
-def convert(mdtex, extensions=[], splitParagraphs=True):
-    ''' converts recursively the Markdown-LaTeX-mixture to HTML with MathML '''
-    found = False
-    # handle all paragraphs separately (prevents aftereffects)
-    if splitParagraphs:
-        parts = re.split("\n\n", mdtex)
-        result = ''
-        for part in parts:
-            result += convert(part, extensions, splitParagraphs=False)
-        return result
-    # find first $$-formula:
-    parts = re.split('\${2}', mdtex, 2)
-    if len(parts)>1:
-        found = True
-        result = convert(parts[0], extensions, splitParagraphs=False)+'\n'
-        try:
-            result += '<div class="blockformula">'+tex2mathml(parts[1])+'</div>\n'
-        except:
-            result += '<div class="blockformula">'+convError+'</div>'
-        if len(parts)==3:
-            result += convert(parts[2], extensions, splitParagraphs=False)
-        else:
-            result += '<div class="blockformula">'+incomplete+'</div>'
-    # else find first $-formulas:
-    else:
-        parts = re.split('\${1}', mdtex, 2)
-    if len(parts)>1 and not found:
-        found = True
-        try:
-            mathml = tex2mathml(parts[1])
-        except:
-            mathml = convError
-        if parts[0].endswith('\n\n') or parts[0]=='': # make sure textblock starts before formula!
-            parts[0]=parts[0]+'&#x200b;'
-        if len(parts)==3:
-            result = convert(parts[0]+mathml+parts[2], extensions, splitParagraphs=False)
-        else:
-            result = convert(parts[0]+mathml+incomplete, extensions, splitParagraphs=False)
-    # else find first \[..\]-equation:
-    else:
-        parts = re.split(r'\\\[', mdtex, 1)
-    if len(parts)>1 and not found:
-        found = True
-        result = convert(parts[0], extensions, splitParagraphs=False)+'\n'
-        parts = re.split(r'\\\]', parts[1], 1)
-        try:
-            result += '<div class="blockformula">'+tex2mathml(parts[0])+'</div>\n'
-        except:
-            result += '<div class="blockformula">'+convError+'</div>'
-        if len(parts)==2:
-            result += convert(parts[1], extensions, splitParagraphs=False)
-        else:
-            result += '<div class="blockformula">'+incomplete+'</div>'
-    # else find first \(..\)-equation:
-    else:
-        parts = re.split(r'\\\(', mdtex, 1)
-    if len(parts)>1 and not found:
-        found = True
-        subp = re.split(r'\\\)', parts[1], 1)
-        try:
-            mathml = tex2mathml(subp[0])
-        except:
-            mathml = convError
-        if parts[0].endswith('\n\n') or parts[0]=='': # make sure textblock starts before formula!
-            parts[0]=parts[0]+'&#x200b;'
-        if len(subp)==2:
-            result = convert(parts[0]+mathml+subp[1], extensions, splitParagraphs=False)
-        else:
-            result = convert(parts[0]+mathml+incomplete, extensions, splitParagraphs=False)
-    if not found:
-        result = mdtex
-    return result

theme.py

@@ -0,0 +1,231 @@
+import gradio as gr
+from toolbox import get_conf
+CODE_HIGHLIGHT, = get_conf('CODE_HIGHLIGHT')
+# gradio可用颜色列表
+# gr.themes.utils.colors.slate (石板色)
+# gr.themes.utils.colors.gray (灰色)
+# gr.themes.utils.colors.zinc (锌色)
+# gr.themes.utils.colors.neutral (中性色)
+# gr.themes.utils.colors.stone (石头色)
+# gr.themes.utils.colors.red (红色)
+# gr.themes.utils.colors.orange (橙色)
+# gr.themes.utils.colors.amber (琥珀色)
+# gr.themes.utils.colors.yellow (黄色)
+# gr.themes.utils.colors.lime (酸橙色)
+# gr.themes.utils.colors.green (绿色)
+# gr.themes.utils.colors.emerald (祖母绿)
+# gr.themes.utils.colors.teal (青蓝色)
+# gr.themes.utils.colors.cyan (青色)
+# gr.themes.utils.colors.sky (天蓝色)
+# gr.themes.utils.colors.blue (蓝色)
+# gr.themes.utils.colors.indigo (靛蓝色)
+# gr.themes.utils.colors.violet (紫罗兰色)
+# gr.themes.utils.colors.purple (紫色)
+# gr.themes.utils.colors.fuchsia (洋红色)
+# gr.themes.utils.colors.pink (粉红色)
+# gr.themes.utils.colors.rose (玫瑰色)
+
+
+def adjust_theme():
+    try:
+        color_er = gr.themes.utils.colors.fuchsia
+        set_theme = gr.themes.Default(
+            primary_hue=gr.themes.utils.colors.orange,
+            neutral_hue=gr.themes.utils.colors.gray,
+            font=["sans-serif", "Microsoft YaHei", "ui-sans-serif", "system-ui",
+                  "sans-serif", gr.themes.utils.fonts.GoogleFont("Source Sans Pro")],
+            font_mono=["ui-monospace", "Consolas", "monospace", gr.themes.utils.fonts.GoogleFont("IBM Plex Mono")])
+        set_theme.set(
+            # Colors
+            input_background_fill_dark="*neutral_800",
+            # Transition
+            button_transition="none",
+            # Shadows
+            button_shadow="*shadow_drop",
+            button_shadow_hover="*shadow_drop_lg",
+            button_shadow_active="*shadow_inset",
+            input_shadow="0 0 0 *shadow_spread transparent, *shadow_inset",
+            input_shadow_focus="0 0 0 *shadow_spread *secondary_50, *shadow_inset",
+            input_shadow_focus_dark="0 0 0 *shadow_spread *neutral_700, *shadow_inset",
+            checkbox_label_shadow="*shadow_drop",
+            block_shadow="*shadow_drop",
+            form_gap_width="1px",
+            # Button borders
+            input_border_width="1px",
+            input_background_fill="white",
+            # Gradients
+            stat_background_fill="linear-gradient(to right, *primary_400, *primary_200)",
+            stat_background_fill_dark="linear-gradient(to right, *primary_400, *primary_600)",
+            error_background_fill=f"linear-gradient(to right, {color_er.c100}, *background_fill_secondary)",
+            error_background_fill_dark="*background_fill_primary",
+            checkbox_label_background_fill="linear-gradient(to top, *neutral_50, white)",
+            checkbox_label_background_fill_dark="linear-gradient(to top, *neutral_900, *neutral_800)",
+            checkbox_label_background_fill_hover="linear-gradient(to top, *neutral_100, white)",
+            checkbox_label_background_fill_hover_dark="linear-gradient(to top, *neutral_900, *neutral_800)",
+            button_primary_background_fill="linear-gradient(to bottom right, *primary_100, *primary_300)",
+            button_primary_background_fill_dark="linear-gradient(to bottom right, *primary_500, *primary_600)",
+            button_primary_background_fill_hover="linear-gradient(to bottom right, *primary_100, *primary_200)",
+            button_primary_background_fill_hover_dark="linear-gradient(to bottom right, *primary_500, *primary_500)",
+            button_primary_border_color_dark="*primary_500",
+            button_secondary_background_fill="linear-gradient(to bottom right, *neutral_100, *neutral_200)",
+            button_secondary_background_fill_dark="linear-gradient(to bottom right, *neutral_600, *neutral_700)",
+            button_secondary_background_fill_hover="linear-gradient(to bottom right, *neutral_100, *neutral_100)",
+            button_secondary_background_fill_hover_dark="linear-gradient(to bottom right, *neutral_600, *neutral_600)",
+            button_cancel_background_fill=f"linear-gradient(to bottom right, {color_er.c100}, {color_er.c200})",
+            button_cancel_background_fill_dark=f"linear-gradient(to bottom right, {color_er.c600}, {color_er.c700})",
+            button_cancel_background_fill_hover=f"linear-gradient(to bottom right, {color_er.c100}, {color_er.c100})",
+            button_cancel_background_fill_hover_dark=f"linear-gradient(to bottom right, {color_er.c600}, {color_er.c600})",
+            button_cancel_border_color=color_er.c200,
+            button_cancel_border_color_dark=color_er.c600,
+            button_cancel_text_color=color_er.c600,
+            button_cancel_text_color_dark="white",
+        )
+    except:
+        set_theme = None
+        print('gradio版本较旧, 不能自定义字体和颜色')
+    return set_theme
+
+
+advanced_css = """
+/* 设置表格的外边距为1em，内部单元格之间边框合并，空单元格显示. */
+.markdown-body table {
+    margin: 1em 0;
+    border-collapse: collapse;
+    empty-cells: show;
+}
+
+/* 设置表格单元格的内边距为5px，边框粗细为1.2px，颜色为--border-color-primary. */
+.markdown-body th, .markdown-body td {
+    border: 1.2px solid var(--border-color-primary);
+    padding: 5px;
+}
+
+/* 设置表头背景颜色为rgba(175,184,193,0.2)，透明度为0.2. */
+.markdown-body thead {
+    background-color: rgba(175,184,193,0.2);
+}
+
+/* 设置表头单元格的内边距为0.5em和0.2em. */
+.markdown-body thead th {
+    padding: .5em .2em;
+}
+
+/* 去掉列表前缀的默认间距，使其与文本线对齐. */
+.markdown-body ol, .markdown-body ul {
+    padding-inline-start: 2em !important;
+}
+
+/* 设定聊天气泡的样式，包括圆角、最大宽度和阴影等. */
+[class *= "message"] {
+    border-radius: var(--radius-xl) !important;
+    /* padding: var(--spacing-xl) !important; */
+    /* font-size: var(--text-md) !important; */
+    /* line-height: var(--line-md) !important; */
+    /* min-height: calc(var(--text-md)*var(--line-md) + 2*var(--spacing-xl)); */
+    /* min-width: calc(var(--text-md)*var(--line-md) + 2*var(--spacing-xl)); */
+}
+[data-testid = "bot"] {
+    max-width: 95%;
+    /* width: auto !important; */
+    border-bottom-left-radius: 0 !important;
+}
+[data-testid = "user"] {
+    max-width: 100%;
+    /* width: auto !important; */
+    border-bottom-right-radius: 0 !important;
+}
+
+/* 行内代码的背景设为淡灰色，设定圆角和间距. */
+.markdown-body code {
+    display: inline;
+    white-space: break-spaces;
+    border-radius: 6px;
+    margin: 0 2px 0 2px;
+    padding: .2em .4em .1em .4em;
+    background-color: rgba(175,184,193,0.2);
+}
+/* 设定代码块的样式，包括背景颜色、内、外边距、圆角。 */
+.markdown-body pre code {
+    display: block;
+    overflow: auto;
+    white-space: pre;
+    background-color: rgba(175,184,193,0.2);
+    border-radius: 10px;
+    padding: 1em;
+    margin: 1em 2em 1em 0.5em;
+}
+
+"""
+
+if CODE_HIGHLIGHT:
+    advanced_css += """
+
+.hll { background-color: #ffffcc }
+.c { color: #3D7B7B; font-style: italic } /* Comment */
+.err { border: 1px solid #FF0000 } /* Error */
+.k { color: hsl(197, 94%, 51%); font-weight: bold } /* Keyword */
+.o { color: #666666 } /* Operator */
+.ch { color: #3D7B7B; font-style: italic } /* Comment.Hashbang */
+.cm { color: #3D7B7B; font-style: italic } /* Comment.Multiline */
+.cp { color: #9C6500 } /* Comment.Preproc */
+.cpf { color: #3D7B7B; font-style: italic } /* Comment.PreprocFile */
+.c1 { color: #3D7B7B; font-style: italic } /* Comment.Single */
+.cs { color: #3D7B7B; font-style: italic } /* Comment.Special */
+.gd { color: #A00000 } /* Generic.Deleted */
+.ge { font-style: italic } /* Generic.Emph */
+.gr { color: #E40000 } /* Generic.Error */
+.gh { color: #000080; font-weight: bold } /* Generic.Heading */
+.gi { color: #008400 } /* Generic.Inserted */
+.go { color: #717171 } /* Generic.Output */
+.gp { color: #000080; font-weight: bold } /* Generic.Prompt */
+.gs { font-weight: bold } /* Generic.Strong */
+.gu { color: #800080; font-weight: bold } /* Generic.Subheading */
+.gt { color: #a9dd00 } /* Generic.Traceback */
+.kc { color: #008000; font-weight: bold } /* Keyword.Constant */
+.kd { color: #008000; font-weight: bold } /* Keyword.Declaration */
+.kn { color: #008000; font-weight: bold } /* Keyword.Namespace */
+.kp { color: #008000 } /* Keyword.Pseudo */
+.kr { color: #008000; font-weight: bold } /* Keyword.Reserved */
+.kt { color: #B00040 } /* Keyword.Type */
+.m { color: #666666 } /* Literal.Number */
+.s { color: #BA2121 } /* Literal.String */
+.na { color: #687822 } /* Name.Attribute */
+.nb { color: #e5f8c3 } /* Name.Builtin */
+.nc { color: #ffad65; font-weight: bold } /* Name.Class */
+.no { color: #880000 } /* Name.Constant */
+.nd { color: #AA22FF } /* Name.Decorator */
+.ni { color: #717171; font-weight: bold } /* Name.Entity */
+.ne { color: #CB3F38; font-weight: bold } /* Name.Exception */
+.nf { color: #f9f978 } /* Name.Function */
+.nl { color: #767600 } /* Name.Label */
+.nn { color: #0000FF; font-weight: bold } /* Name.Namespace */
+.nt { color: #008000; font-weight: bold } /* Name.Tag */
+.nv { color: #19177C } /* Name.Variable */
+.ow { color: #AA22FF; font-weight: bold } /* Operator.Word */
+.w { color: #bbbbbb } /* Text.Whitespace */
+.mb { color: #666666 } /* Literal.Number.Bin */
+.mf { color: #666666 } /* Literal.Number.Float */
+.mh { color: #666666 } /* Literal.Number.Hex */
+.mi { color: #666666 } /* Literal.Number.Integer */
+.mo { color: #666666 } /* Literal.Number.Oct */
+.sa { color: #BA2121 } /* Literal.String.Affix */
+.sb { color: #BA2121 } /* Literal.String.Backtick */
+.sc { color: #BA2121 } /* Literal.String.Char */
+.dl { color: #BA2121 } /* Literal.String.Delimiter */
+.sd { color: #BA2121; font-style: italic } /* Literal.String.Doc */
+.s2 { color: #2bf840 } /* Literal.String.Double */
+.se { color: #AA5D1F; font-weight: bold } /* Literal.String.Escape */
+.sh { color: #BA2121 } /* Literal.String.Heredoc */
+.si { color: #A45A77; font-weight: bold } /* Literal.String.Interpol */
+.sx { color: #008000 } /* Literal.String.Other */
+.sr { color: #A45A77 } /* Literal.String.Regex */
+.s1 { color: #BA2121 } /* Literal.String.Single */
+.ss { color: #19177C } /* Literal.String.Symbol */
+.bp { color: #008000 } /* Name.Builtin.Pseudo */
+.fm { color: #0000FF } /* Name.Function.Magic */
+.vc { color: #19177C } /* Name.Variable.Class */
+.vg { color: #19177C } /* Name.Variable.Global */
+.vi { color: #19177C } /* Name.Variable.Instance */
+.vm { color: #19177C } /* Name.Variable.Magic */
+.il { color: #666666 } /* Literal.Number.Integer.Long */
+"""

toolbox.py

@@ -0,0 +1,529 @@
+import markdown
+import mdtex2html
+import threading
+import importlib
+import traceback
+import inspect
+import re
+from latex2mathml.converter import convert as tex2mathml
+from functools import wraps, lru_cache
+
+############################### 插件输入输出接驳区 #######################################
+class ChatBotWithCookies(list):
+    def __init__(self, cookie):
+        self._cookies = cookie
+
+    def write_list(self, list):
+        for t in list:
+            self.append(t)
+
+    def get_list(self):
+        return [t for t in self]
+
+    def get_cookies(self):
+        return self._cookies
+
+def ArgsGeneralWrapper(f):
+    """
+        装饰器函数，用于重组输入参数，改变输入参数的顺序与结构。
+    """
+    def decorated(cookies, txt, txt2, top_p, temperature, chatbot, history, system_prompt, *args):
+        txt_passon = txt
+        if txt == "" and txt2 != "": txt_passon = txt2
+        # 引入一个有cookie的chatbot
+        cookies.update({
+            'top_p':top_p, 
+            'temperature':temperature,
+        })
+        llm_kwargs = {
+            'api_key': cookies['api_key'],
+            'llm_model': cookies['llm_model'],
+            'top_p':top_p, 
+            'temperature':temperature,
+        }
+        plugin_kwargs = {
+            # 目前还没有
+        }
+        chatbot_with_cookie = ChatBotWithCookies(cookies)
+        chatbot_with_cookie.write_list(chatbot)
+        yield from f(txt_passon, llm_kwargs, plugin_kwargs, chatbot_with_cookie, history, system_prompt, *args)
+    return decorated
+
+def update_ui(chatbot, history, msg='正常', **kwargs):  # 刷新界面
+    """
+    刷新用户界面
+    """
+    assert isinstance(chatbot, ChatBotWithCookies), "在传递chatbot的过程中不要将其丢弃。必要时，可用clear将其清空，然后用for+append循环重新赋值。"
+    yield chatbot.get_cookies(), chatbot, history, msg
+############################### ################## #######################################
+##########################################################################################
+
+def get_reduce_token_percent(text):
+    """
+        * 此函数未来将被弃用
+    """
+    try:
+        # text = "maximum context length is 4097 tokens. However, your messages resulted in 4870 tokens"
+        pattern = r"(\d+)\s+tokens\b"
+        match = re.findall(pattern, text)
+        EXCEED_ALLO = 500  # 稍微留一点余地，否则在回复时会因余量太少出问题
+        max_limit = float(match[0]) - EXCEED_ALLO
+        current_tokens = float(match[1])
+        ratio = max_limit/current_tokens
+        assert ratio > 0 and ratio < 1
+        return ratio, str(int(current_tokens-max_limit))
+    except:
+        return 0.5, '不详'
+
+def predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, llm_kwargs, history=[], sys_prompt='', long_connection=True):
+    """
+        * 此函数未来将被弃用（替代函数 request_gpt_model_in_new_thread_with_ui_alive 文件 chatgpt_academic/crazy_functions/crazy_utils）
+
+        调用简单的predict_no_ui接口，但是依然保留了些许界面心跳功能，当对话太长时，会自动采用二分法截断
+        i_say: 当前输入
+        i_say_show_user: 显示到对话界面上的当前输入，例如，输入整个文件时，你绝对不想把文件的内容都糊到对话界面上
+        chatbot: 对话界面句柄
+        top_p, temperature: gpt参数
+        history: gpt参数 对话历史
+        sys_prompt: gpt参数 sys_prompt
+        long_connection: 是否采用更稳定的连接方式（推荐）（已弃用）
+    """
+    import time
+    from request_llm.bridge_chatgpt import predict_no_ui_long_connection
+    from toolbox import get_conf
+    TIMEOUT_SECONDS, MAX_RETRY = get_conf('TIMEOUT_SECONDS', 'MAX_RETRY')
+    # 多线程的时候，需要一个mutable结构在不同线程之间传递信息
+    # list就是最简单的mutable结构，我们第一个位置放gpt输出，第二个位置传递报错信息
+    mutable = [None, '']
+    # multi-threading worker
+
+    def mt(i_say, history):
+        while True:
+            try:
+                mutable[0] = predict_no_ui_long_connection(
+                    inputs=i_say, llm_kwargs=llm_kwargs, history=history, sys_prompt=sys_prompt)
+
+            except ConnectionAbortedError as token_exceeded_error:
+                # 尝试计算比例，尽可能多地保留文本
+                p_ratio, n_exceed = get_reduce_token_percent(
+                    str(token_exceeded_error))
+                if len(history) > 0:
+                    history = [his[int(len(his) * p_ratio):]
+                               for his in history if his is not None]
+                else:
+                    i_say = i_say[:     int(len(i_say) * p_ratio)]
+                mutable[1] = f'警告，文本过长将进行截断，Token溢出数：{n_exceed}，截断比例：{(1-p_ratio):.0%}。'
+            except TimeoutError as e:
+                mutable[0] = '[Local Message] 请求超时。'
+                raise TimeoutError
+            except Exception as e:
+                mutable[0] = f'[Local Message] 异常：{str(e)}.'
+                raise RuntimeError(f'[Local Message] 异常：{str(e)}.')
+    # 创建新线程发出http请求
+    thread_name = threading.Thread(target=mt, args=(i_say, history))
+    thread_name.start()
+    # 原来的线程则负责持续更新UI，实现一个超时倒计时，并等待新线程的任务完成
+    cnt = 0
+    while thread_name.is_alive():
+        cnt += 1
+        chatbot[-1] = (i_say_show_user,
+                       f"[Local Message] {mutable[1]}waiting gpt response {cnt}/{TIMEOUT_SECONDS*2*(MAX_RETRY+1)}"+''.join(['.']*(cnt % 4)))
+        yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
+        time.sleep(1)
+    # 把gpt的输出从mutable中取出来
+    gpt_say = mutable[0]
+    if gpt_say == '[Local Message] Failed with timeout.':
+        raise TimeoutError
+    return gpt_say
+
+
+def write_results_to_file(history, file_name=None):
+    """
+        将对话记录history以Markdown格式写入文件中。如果没有指定文件名，则使用当前时间生成文件名。
+    """
+    import os
+    import time
+    if file_name is None:
+        # file_name = time.strftime("chatGPT分析报告%Y-%m-%d-%H-%M-%S", time.localtime()) + '.md'
+        file_name = 'chatGPT分析报告' + \
+            time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) + '.md'
+    os.makedirs('./gpt_log/', exist_ok=True)
+    with open(f'./gpt_log/{file_name}', 'w', encoding='utf8') as f:
+        f.write('# chatGPT 分析报告\n')
+        for i, content in enumerate(history):
+            try:    # 这个bug没找到触发条件，暂时先这样顶一下
+                if type(content) != str:
+                    content = str(content)
+            except:
+                continue
+            if i % 2 == 0:
+                f.write('## ')
+            f.write(content)
+            f.write('\n\n')
+    res = '以上材料已经被写入' + os.path.abspath(f'./gpt_log/{file_name}')
+    print(res)
+    return res
+
+
+def regular_txt_to_markdown(text):
+    """
+        将普通文本转换为Markdown格式的文本。
+    """
+    text = text.replace('\n', '\n\n')
+    text = text.replace('\n\n\n', '\n\n')
+    text = text.replace('\n\n\n', '\n\n')
+    return text
+
+
+def CatchException(f):
+    """
+        装饰器函数，捕捉函数f中的异常并封装到一个生成器中返回，并显示到聊天当中。
+    """
+    @wraps(f)
+    def decorated(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+        try:
+            yield from f(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT)
+        except Exception as e:
+            from check_proxy import check_proxy
+            from toolbox import get_conf
+            proxies, = get_conf('proxies')
+            tb_str = '```\n' + traceback.format_exc() + '```'
+            if chatbot is None or len(chatbot) == 0:
+                chatbot = [["插件调度异常", "异常原因"]]
+            chatbot[-1] = (chatbot[-1][0],
+                           f"[Local Message] 实验性函数调用出错: \n\n{tb_str} \n\n当前代理可用性: \n\n{check_proxy(proxies)}")
+            yield from update_ui(chatbot=chatbot, history=history, msg=f'异常 {e}') # 刷新界面
+    return decorated
+
+
+def HotReload(f):
+    """
+    HotReload的装饰器函数，用于实现Python函数插件的热更新。
+    函数热更新是指在不停止程序运行的情况下，更新函数代码，从而达到实时更新功能。
+    在装饰器内部，使用wraps(f)来保留函数的元信息，并定义了一个名为decorated的内部函数。
+    内部函数通过使用importlib模块的reload函数和inspect模块的getmodule函数来重新加载并获取函数模块，
+    然后通过getattr函数获取函数名，并在新模块中重新加载函数。
+    最后，使用yield from语句返回重新加载过的函数，并在被装饰的函数上执行。
+    最终，装饰器函数返回内部函数。这个内部函数可以将函数的原始定义更新为最新版本，并执行函数的新版本。
+    """
+    @wraps(f)
+    def decorated(*args, **kwargs):
+        fn_name = f.__name__
+        f_hot_reload = getattr(importlib.reload(inspect.getmodule(f)), fn_name)
+        yield from f_hot_reload(*args, **kwargs)
+    return decorated
+
+
+def report_execption(chatbot, history, a, b):
+    """
+        向chatbot中添加错误信息
+    """
+    chatbot.append((a, b))
+    history.append(a)
+    history.append(b)
+
+
+def text_divide_paragraph(text):
+    """
+        将文本按照段落分隔符分割开，生成带有段落标签的HTML代码。
+    """
+    if '```' in text:
+        # careful input
+        return text
+    else:
+        # wtf input
+        lines = text.split("\n")
+        for i, line in enumerate(lines):
+            lines[i] = lines[i].replace(" ", "&nbsp;")
+        text = "</br>".join(lines)
+        return text
+
+
+def markdown_convertion(txt):
+    """
+        将Markdown格式的文本转换为HTML格式。如果包含数学公式，则先将公式转换为HTML格式。
+    """
+    pre = '<div class="markdown-body">'
+    suf = '</div>'
+    markdown_extension_configs = {
+        'mdx_math': {
+            'enable_dollar_delimiter': True,
+            'use_gitlab_delimiters': False,
+        },
+    }
+    find_equation_pattern = r'<script type="math/tex(?:.*?)>(.*?)</script>'
+
+    def tex2mathml_catch_exception(content, *args, **kwargs):
+        try:
+            content = tex2mathml(content, *args, **kwargs)
+        except:
+            content = content
+        return content
+
+    def replace_math_no_render(match):
+        content = match.group(1)
+        if 'mode=display' in match.group(0):
+            content = content.replace('\n', '</br>')
+            return f"<font color=\"#00FF00\">$$</font><font color=\"#FF00FF\">{content}</font><font color=\"#00FF00\">$$</font>"
+        else:
+            return f"<font color=\"#00FF00\">$</font><font color=\"#FF00FF\">{content}</font><font color=\"#00FF00\">$</font>"
+
+    def replace_math_render(match):
+        content = match.group(1)
+        if 'mode=display' in match.group(0):
+            if '\\begin{aligned}' in content:
+                content = content.replace('\\begin{aligned}', '\\begin{array}')
+                content = content.replace('\\end{aligned}', '\\end{array}')
+                content = content.replace('&', ' ')
+            content = tex2mathml_catch_exception(content, display="block")
+            return content
+        else:
+            return tex2mathml_catch_exception(content)
+        
+    def markdown_bug_hunt(content):
+        """
+        解决一个mdx_math的bug（单$包裹begin命令时多余<script>）
+        """
+        content = content.replace('<script type="math/tex">\n<script type="math/tex; mode=display">', '<script type="math/tex; mode=display">')
+        content = content.replace('</script>\n</script>', '</script>')
+        return content
+    
+
+    if ('$' in txt) and ('```' not in txt):  # 有$标识的公式符号，且没有代码段```的标识
+        # convert everything to html format
+        split = markdown.markdown(text='---')
+        convert_stage_1 = markdown.markdown(text=txt, extensions=['mdx_math', 'fenced_code', 'tables', 'sane_lists'], extension_configs=markdown_extension_configs)
+        convert_stage_1 = markdown_bug_hunt(convert_stage_1)
+        # re.DOTALL: Make the '.' special character match any character at all, including a newline; without this flag, '.' will match anything except a newline. Corresponds to the inline flag (?s).
+        # 1. convert to easy-to-copy tex (do not render math)
+        convert_stage_2_1, n = re.subn(find_equation_pattern, replace_math_no_render, convert_stage_1, flags=re.DOTALL)
+        # 2. convert to rendered equation
+        convert_stage_2_2, n = re.subn(find_equation_pattern, replace_math_render, convert_stage_1, flags=re.DOTALL)
+        # cat them together
+        return pre + convert_stage_2_1 + f'{split}' + convert_stage_2_2 + suf
+    else:
+        return pre + markdown.markdown(txt, extensions=['fenced_code', 'codehilite', 'tables', 'sane_lists']) + suf
+
+
+def close_up_code_segment_during_stream(gpt_reply):
+    """
+    在gpt输出代码的中途（输出了前面的```，但还没输出完后面的```），补上后面的```
+    
+    Args:
+        gpt_reply (str): GPT模型返回的回复字符串。
+
+    Returns:
+        str: 返回一个新的字符串，将输出代码片段的“后面的```”补上。
+
+    """
+    if '```' not in gpt_reply:
+        return gpt_reply
+    if gpt_reply.endswith('```'):
+        return gpt_reply
+
+    # 排除了以上两个情况，我们
+    segments = gpt_reply.split('```')
+    n_mark = len(segments) - 1
+    if n_mark % 2 == 1:
+        # print('输出代码片段中！')
+        return gpt_reply+'\n```'
+    else:
+        return gpt_reply
+
+
+def format_io(self, y):
+    """
+        将输入和输出解析为HTML格式。将y中最后一项的输入部分段落化，并将输出部分的Markdown和数学公式转换为HTML格式。
+    """
+    if y is None or y == []:
+        return []
+    i_ask, gpt_reply = y[-1]
+    i_ask = text_divide_paragraph(i_ask)  # 输入部分太自由，预处理一波
+    gpt_reply = close_up_code_segment_during_stream(gpt_reply)  # 当代码输出半截的时候，试着补上后个```
+    y[-1] = (
+        None if i_ask is None else markdown.markdown(i_ask, extensions=['fenced_code', 'tables']),
+        None if gpt_reply is None else markdown_convertion(gpt_reply)
+    )
+    return y
+
+
+def find_free_port():
+    """
+        返回当前系统中可用的未使用端口。
+    """
+    import socket
+    from contextlib import closing
+    with closing(socket.socket(socket.AF_INET, socket.SOCK_STREAM)) as s:
+        s.bind(('', 0))
+        s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+        return s.getsockname()[1]
+
+
+def extract_archive(file_path, dest_dir):
+    import zipfile
+    import tarfile
+    import os
+    # Get the file extension of the input file
+    file_extension = os.path.splitext(file_path)[1]
+
+    # Extract the archive based on its extension
+    if file_extension == '.zip':
+        with zipfile.ZipFile(file_path, 'r') as zipobj:
+            zipobj.extractall(path=dest_dir)
+            print("Successfully extracted zip archive to {}".format(dest_dir))
+
+    elif file_extension in ['.tar', '.gz', '.bz2']:
+        with tarfile.open(file_path, 'r:*') as tarobj:
+            tarobj.extractall(path=dest_dir)
+            print("Successfully extracted tar archive to {}".format(dest_dir))
+
+    # 第三方库，需要预先pip install rarfile
+    # 此外，Windows上还需要安装winrar软件，配置其Path环境变量，如"C:\Program Files\WinRAR"才可以
+    elif file_extension == '.rar':
+        try:
+            import rarfile
+            with rarfile.RarFile(file_path) as rf:
+                rf.extractall(path=dest_dir)
+                print("Successfully extracted rar archive to {}".format(dest_dir))
+        except:
+            print("Rar format requires additional dependencies to install")
+            return '\n\n需要安装pip install rarfile来解压rar文件'
+
+    # 第三方库，需要预先pip install py7zr
+    elif file_extension == '.7z':
+        try:
+            import py7zr
+            with py7zr.SevenZipFile(file_path, mode='r') as f:
+                f.extractall(path=dest_dir)
+                print("Successfully extracted 7z archive to {}".format(dest_dir))
+        except:
+            print("7z format requires additional dependencies to install")
+            return '\n\n需要安装pip install py7zr来解压7z文件'
+    else:
+        return ''
+    return ''
+
+
+def find_recent_files(directory):
+    """
+        me: find files that is created with in one minutes under a directory with python, write a function
+        gpt: here it is!
+    """
+    import os
+    import time
+    current_time = time.time()
+    one_minute_ago = current_time - 60
+    recent_files = []
+
+    for filename in os.listdir(directory):
+        file_path = os.path.join(directory, filename)
+        if file_path.endswith('.log'):
+            continue
+        created_time = os.path.getmtime(file_path)
+        if created_time >= one_minute_ago:
+            if os.path.isdir(file_path):
+                continue
+            recent_files.append(file_path)
+
+    return recent_files
+
+
+def on_file_uploaded(files, chatbot, txt):
+    if len(files) == 0:
+        return chatbot, txt
+    import shutil
+    import os
+    import time
+    import glob
+    from toolbox import extract_archive
+    try:
+        shutil.rmtree('./private_upload/')
+    except:
+        pass
+    time_tag = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime())
+    os.makedirs(f'private_upload/{time_tag}', exist_ok=True)
+    err_msg = ''
+    for file in files:
+        file_origin_name = os.path.basename(file.orig_name)
+        shutil.copy(file.name, f'private_upload/{time_tag}/{file_origin_name}')
+        err_msg += extract_archive(f'private_upload/{time_tag}/{file_origin_name}',
+                                   dest_dir=f'private_upload/{time_tag}/{file_origin_name}.extract')
+    moved_files = [fp for fp in glob.glob(
+        'private_upload/**/*', recursive=True)]
+    txt = f'private_upload/{time_tag}'
+    moved_files_str = '\t\n\n'.join(moved_files)
+    chatbot.append(['我上传了文件，请查收',
+                    f'[Local Message] 收到以下文件: \n\n{moved_files_str}' +
+                    f'\n\n调用路径参数已自动修正到: \n\n{txt}' +
+                    f'\n\n现在您点击任意实验功能时，以上文件将被作为输入参数'+err_msg])
+    return chatbot, txt
+
+
+def on_report_generated(files, chatbot):
+    from toolbox import find_recent_files
+    report_files = find_recent_files('gpt_log')
+    if len(report_files) == 0:
+        return None, chatbot
+    # files.extend(report_files)
+    chatbot.append(['汇总报告如何远程获取？', '汇总报告已经添加到右侧“文件上传区”（可能处于折叠状态），请查收。'])
+    return report_files, chatbot
+
+def is_openai_api_key(key):
+    # 正确的 API_KEY 是 "sk-" + 48 位大小写字母数字的组合
+    API_MATCH = re.match(r"sk-[a-zA-Z0-9]{48}$", key)
+    return API_MATCH
+
+@lru_cache(maxsize=128)
+def read_single_conf_with_lru_cache(arg):
+    from colorful import print亮红, print亮绿
+    try:
+        r = getattr(importlib.import_module('config_private'), arg)
+    except:
+        r = getattr(importlib.import_module('config'), arg)
+    # 在读取API_KEY时，检查一下是不是忘了改config
+    if arg == 'API_KEY':
+        if is_openai_api_key(r):
+            print亮绿(f"[API_KEY] 您的 API_KEY 是: {r[:15]}*** API_KEY 导入成功")
+        else:
+            print亮红( "[API_KEY] 正确的 API_KEY 是 'sk-' + '48 位大小写字母数字' 的组合，请在config文件中修改API密钥, 添加海外代理之后再运行。" + \
+                "（如果您刚更新过代码，请确保旧版config_private文件中没有遗留任何新增键值）")
+    if arg == 'proxies':
+        if r is None:
+            print亮红('[PROXY] 网络代理状态：未配置。无代理状态下很可能无法访问。建议：检查USE_PROXY选项是否修改。')
+        else:
+            print亮绿('[PROXY] 网络代理状态：已配置。配置信息如下：', r)
+            assert isinstance(r, dict), 'proxies格式错误，请注意proxies选项的格式，不要遗漏括号。'
+    return r
+
+
+def get_conf(*args):
+    # 建议您复制一个config_private.py放自己的秘密, 如API和代理网址, 避免不小心传github被别人看到
+    res = []
+    for arg in args:
+        r = read_single_conf_with_lru_cache(arg)
+        res.append(r)
+    return res
+
+
+def clear_line_break(txt):
+    txt = txt.replace('\n', ' ')
+    txt = txt.replace('  ', ' ')
+    txt = txt.replace('  ', ' ')
+    return txt
+
+
+class DummyWith():
+    """
+    这段代码定义了一个名为DummyWith的空上下文管理器，
+    它的作用是……额……没用，即在代码结构不变得情况下取代其他的上下文管理器。
+    上下文管理器是一种Python对象，用于与with语句一起使用，
+    以确保一些资源在代码块执行期间得到正确的初始化和清理。
+    上下文管理器必须实现两个方法，分别为 __enter__()和 __exit__()。 
+    在上下文执行开始的情况下，__enter__()方法会在代码块被执行前被调用，
+    而在上下文执行结束时，__exit__()方法则会被调用。
+    """
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc_value, traceback):
+        return

version

@@ -0,0 +1,5 @@
+{
+  "version": 2.68,
+  "show_feature": true,
+  "new_feature": "改善理解pdf（chatpdf）功能 <-> 如果一键更新失败，可前往github手动更新"
+}