diff --git a/.github/workflows/build-with-all-capacity-beta.yml b/.github/workflows/build-with-all-capacity-beta.yml new file mode 100644 index 00000000..5a2a1a54 --- /dev/null +++ b/.github/workflows/build-with-all-capacity-beta.yml @@ -0,0 +1,44 @@ +# https://docs.github.com/en/actions/publishing-packages/publishing-docker-images#publishing-images-to-github-packages +name: build-with-all-capacity-beta + +on: + push: + branches: + - 'master' + +env: + REGISTRY: ghcr.io + IMAGE_NAME: ${{ github.repository }}_with_all_capacity_beta + +jobs: + build-and-push-image: + runs-on: ubuntu-latest + permissions: + contents: read + packages: write + + steps: + - name: Checkout repository + uses: actions/checkout@v3 + + - name: Log in to the Container registry + uses: docker/login-action@v2 + with: + registry: ${{ env.REGISTRY }} + username: ${{ github.actor }} + password: ${{ secrets.GITHUB_TOKEN }} + + - name: Extract metadata (tags, labels) for Docker + id: meta + uses: docker/metadata-action@v4 + with: + images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }} + + - name: Build and push Docker image + uses: docker/build-push-action@v4 + with: + context: . + push: true + file: docs/GithubAction+AllCapacityBeta + tags: ${{ steps.meta.outputs.tags }} + labels: ${{ steps.meta.outputs.labels }} diff --git a/README.md b/README.md index 3167cf9a..922cb229 100644 --- a/README.md +++ b/README.md @@ -14,41 +14,69 @@ pinned: false > > 2023.11.12: 某些依赖包尚不兼容python 3.12，推荐python 3.11。 > -> 2023.11.7: 安装依赖时，请选择`requirements.txt`中**指定的版本**。 安装命令：`pip install -r requirements.txt`。本项目开源免费，近期发现有人蔑视开源协议并利用本项目违规圈钱，请提高警惕，谨防上当受骗。 +> 2023.12.26: 安装依赖时，请选择`requirements.txt`中**指定的版本**。 安装命令：`pip install -r requirements.txt`。本项目完全开源免费，您可通过订阅[在线服务](https://github.com/binary-husky/gpt_academic/wiki/online)的方式鼓励本项目的发展。 + +
+ +
+

+ GPT 学术优化 (GPT Academic) +

+ +[![Github][Github-image]][Github-url] +[![License][License-image]][License-url] +[![Releases][Releases-image]][Releases-url] +[![Installation][Installation-image]][Installation-url] +[![Wiki][Wiki-image]][Wiki-url] +[![PR][PRs-image]][PRs-url] + +[Github-image]: https://img.shields.io/badge/github-12100E.svg?style=flat-square +[License-image]: https://img.shields.io/github/license/binary-husky/gpt_academic?label=License&style=flat-square&color=orange +[Releases-image]: https://img.shields.io/github/release/binary-husky/gpt_academic?label=Release&style=flat-square&color=blue +[Installation-image]: https://img.shields.io/badge/dynamic/json?color=blue&url=https://raw.githubusercontent.com/binary-husky/gpt_academic/master/version&query=$.version&label=Installation&style=flat-square +[Wiki-image]: https://img.shields.io/badge/wiki-项目文档-black?style=flat-square +[PRs-image]: https://img.shields.io/badge/PRs-welcome-pink?style=flat-square + +[Github-url]: https://github.com/binary-husky/gpt_academic +[License-url]: https://github.com/binary-husky/gpt_academic/blob/master/LICENSE +[Releases-url]: https://github.com/binary-husky/gpt_academic/releases +[Installation-url]: https://github.com/binary-husky/gpt_academic#installation +[Wiki-url]: https://github.com/binary-husky/gpt_academic/wiki +[PRs-url]: https://github.com/binary-husky/gpt_academic/pulls - -#
GPT 学术优化 (GPT Academic)
+
+
**如果喜欢这个项目，请给它一个Star；如果您发明了好用的快捷键或插件，欢迎发pull requests！** -If you like this project, please give it a Star. We also have a README in [English|](docs/README.English.md)[日本語|](docs/README.Japanese.md)[한국어|](docs/README.Korean.md)[Русский|](docs/README.Russian.md)[Français](docs/README.French.md) translated by this project itself. -To translate this project to arbitrary language with GPT, read and run [`multi_language.py`](multi_language.py) (experimental). +If you like this project, please give it a Star. +Read this in [English](docs/README.English.md) | [日本語](docs/README.Japanese.md) | [한국어](docs/README.Korean.md) | [Русский](docs/README.Russian.md) | [Français](docs/README.French.md). All translations have been provided by the project itself. To translate this project to arbitrary language with GPT, read and run [`multi_language.py`](multi_language.py) (experimental). +
+ -> **Note** -> > 1.请注意只有 **高亮** 标识的插件（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[自译解报告`self_analysis.md`](https://github.com/binary-husky/gpt_academic/wiki/GPT‐Academic项目自译解报告)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[`wiki`](https://github.com/binary-husky/gpt_academic/wiki)。[常规安装方法](#installation) | [一键安装脚本](https://github.com/binary-husky/gpt_academic/releases) | [配置说明](https://github.com/binary-husky/gpt_academic/wiki/项目配置说明)。 +> 2.本项目中每个文件的功能都在[自译解报告](https://github.com/binary-husky/gpt_academic/wiki/GPT‐Academic项目自译解报告)`self_analysis.md`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。常见问题请查阅wiki。 +> [![常规安装方法](https://img.shields.io/static/v1?label=&message=常规安装方法&color=gray)](#installation) [![一键安装脚本](https://img.shields.io/static/v1?label=&message=一键安装脚本&color=gray)](https://github.com/binary-husky/gpt_academic/releases) [![配置说明](https://img.shields.io/static/v1?label=&message=配置说明&color=gray)](https://github.com/binary-husky/gpt_academic/wiki/项目配置说明) [![wiki](https://img.shields.io/static/v1?label=&message=wiki&color=gray)]([https://github.com/binary-husky/gpt_academic/wiki/项目配置说明](https://github.com/binary-husky/gpt_academic/wiki)) > -> 3.本项目兼容并鼓励尝试国产大语言模型ChatGLM等。支持多个api-key共存，可在配置文件中填写如`API_KEY="openai-key1,openai-key2,azure-key3,api2d-key4"`。需要临时更换`API_KEY`时，在输入区输入临时的`API_KEY`然后回车键提交后即可生效。 +> 3.本项目兼容并鼓励尝试国产大语言模型ChatGLM等。支持多个api-key共存，可在配置文件中填写如`API_KEY="openai-key1,openai-key2,azure-key3,api2d-key4"`。需要临时更换`API_KEY`时，在输入区输入临时的`API_KEY`然后回车键提交即可生效。 - - +

功能（⭐= 近期新增功能） | 描述 --- | --- -⭐[接入新模型](https://github.com/binary-husky/gpt_academic/wiki/%E5%A6%82%E4%BD%95%E5%88%87%E6%8D%A2%E6%A8%A1%E5%9E%8B)！ | 百度[千帆](https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Nlks5zkzu)与文心一言, 通义千问[Qwen](https://modelscope.cn/models/qwen/Qwen-7B-Chat/summary)，上海AI-Lab[书生](https://github.com/InternLM/InternLM)，讯飞[星火](https://xinghuo.xfyun.cn/)，[LLaMa2](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf)，[智谱API](https://open.bigmodel.cn/)，DALLE3, [DeepseekCoder](https://coder.deepseek.com/) +⭐[接入新模型](https://github.com/binary-husky/gpt_academic/wiki/%E5%A6%82%E4%BD%95%E5%88%87%E6%8D%A2%E6%A8%A1%E5%9E%8B) | 百度[千帆](https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Nlks5zkzu)与文心一言, 通义千问[Qwen](https://modelscope.cn/models/qwen/Qwen-7B-Chat/summary)，上海AI-Lab[书生](https://github.com/InternLM/InternLM)，讯飞[星火](https://xinghuo.xfyun.cn/)，[LLaMa2](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf)，[智谱API](https://open.bigmodel.cn/)，DALLE3, [DeepseekCoder](https://coder.deepseek.com/) 润色、翻译、代码解释 | 一键润色、翻译、查找论文语法错误、解释代码 [自定义快捷键](https://www.bilibili.com/video/BV14s4y1E7jN) | 支持自定义快捷键 模块化设计 | 支持自定义强大的[插件](https://github.com/binary-husky/gpt_academic/tree/master/crazy_functions)，插件支持[热更新](https://github.com/binary-husky/gpt_academic/wiki/%E5%87%BD%E6%95%B0%E6%8F%92%E4%BB%B6%E6%8C%87%E5%8D%97) -[程序剖析](https://www.bilibili.com/video/BV1cj411A7VW) | [插件] 一键可以剖析Python/C/C++/Java/Lua/...项目树 或 [自我剖析](https://www.bilibili.com/video/BV1cj411A7VW) +[程序剖析](https://www.bilibili.com/video/BV1cj411A7VW) | [插件] 一键剖析Python/C/C++/Java/Lua/...项目树 或 [自我剖析](https://www.bilibili.com/video/BV1cj411A7VW) 读论文、[翻译](https://www.bilibili.com/video/BV1KT411x7Wn)论文 | [插件] 一键解读latex/pdf论文全文并生成摘要 Latex全文[翻译](https://www.bilibili.com/video/BV1nk4y1Y7Js/)、[润色](https://www.bilibili.com/video/BV1FT411H7c5/) | [插件] 一键翻译或润色latex论文 批量注释生成 | [插件] 一键批量生成函数注释 -Markdown[中英互译](https://www.bilibili.com/video/BV1yo4y157jV/) | [插件] 看到上面5种语言的[README](https://github.com/binary-husky/gpt_academic/blob/master/docs/README_EN.md)了吗？ +Markdown[中英互译](https://www.bilibili.com/video/BV1yo4y157jV/) | [插件] 看到上面5种语言的[README](https://github.com/binary-husky/gpt_academic/blob/master/docs/README_EN.md)了吗？就是出自他的手笔 chat分析报告生成 | [插件] 运行后自动生成总结汇报 [PDF论文全文翻译功能](https://www.bilibili.com/video/BV1KT411x7Wn) | [插件] PDF论文提取题目&摘要+翻译全文（多线程） [Arxiv小助手](https://www.bilibili.com/video/BV1LM4y1279X) | [插件] 输入arxiv文章url即可一键翻译摘要+下载PDF @@ -60,22 +88,22 @@ Latex论文一键校对 | [插件] 仿Grammarly对Latex文章进行语法、拼 公式/图片/表格显示 | 可以同时显示公式的[tex形式和渲染形式](https://user-images.githubusercontent.com/96192199/230598842-1d7fcddd-815d-40ee-af60-baf488a199df.png)，支持公式、代码高亮 ⭐AutoGen多智能体插件 | [插件] 借助微软AutoGen，探索多Agent的智能涌现可能！ 启动暗色[主题](https://github.com/binary-husky/gpt_academic/issues/173) | 在浏览器url后面添加```/?__theme=dark```可以切换dark主题 -[多LLM模型](https://www.bilibili.com/video/BV1wT411p7yf)支持 | 同时被GPT3.5、GPT4、[清华ChatGLM2](https://github.com/THUDM/ChatGLM2-6B)、[复旦MOSS](https://github.com/OpenLMLab/MOSS)同时伺候的感觉一定会很不错吧？ +[多LLM模型](https://www.bilibili.com/video/BV1wT411p7yf)支持 | 同时被GPT3.5、GPT4、[清华ChatGLM2](https://github.com/THUDM/ChatGLM2-6B)、[复旦MOSS](https://github.com/OpenLMLab/MOSS)伺候的感觉一定会很不错吧？ ⭐ChatGLM2微调模型 | 支持加载ChatGLM2微调模型，提供ChatGLM2微调辅助插件 更多LLM模型接入，支持[huggingface部署](https://huggingface.co/spaces/qingxu98/gpt-academic) | 加入Newbing接口(新必应)，引入清华[Jittorllms](https://github.com/Jittor/JittorLLMs)支持[LLaMA](https://github.com/facebookresearch/llama)和[盘古α](https://openi.org.cn/pangu/) ⭐[void-terminal](https://github.com/binary-husky/void-terminal) pip包 | 脱离GUI，在Python中直接调用本项目的所有函数插件（开发中） -⭐虚空终端插件 | [插件] 用自然语言，直接调度本项目其他插件 +⭐虚空终端插件 | [插件] 能够使用自然语言直接调度本项目其他插件 更多新功能展示 (图像生成等) …… | 见本文档结尾处 ……
- 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）
- +
-- 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 +- 所有按钮都通过读取functional.py动态生成，可随意加自定义功能，解放剪贴板
@@ -85,21 +113,23 @@ Latex论文一键校对 | [插件] 仿Grammarly对Latex文章进行语法、拼 -- 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 +- 如果输出包含公式，会以tex形式和渲染形式同时显示，方便复制和阅读
-- 懒得看项目代码？整个工程直接给chatgpt炫嘴里 +- 懒得看项目代码？直接把整个工程炫ChatGPT嘴里
-- 多种大语言模型混合调用（ChatGLM + OpenAI-GPT3.5 + [API2D](https://api2d.com/)-GPT4） +- 多种大语言模型混合调用（ChatGLM + OpenAI-GPT3.5 + GPT4）
+

+ # Installation ### 安装方法I：直接运行 (Windows, Linux or MacOS) @@ -110,13 +140,13 @@ Latex论文一键校对 | [插件] 仿Grammarly对Latex文章进行语法、拼 cd gpt_academic ``` -2. 配置API_KEY +2. 配置API_KEY等变量 - 在`config.py`中，配置API KEY等设置，[点击查看特殊网络环境设置方法](https://github.com/binary-husky/gpt_academic/issues/1) 。[Wiki页面](https://github.com/binary-husky/gpt_academic/wiki/项目配置说明)。 + 在`config.py`中，配置API KEY等变量。[特殊网络环境设置方法](https://github.com/binary-husky/gpt_academic/issues/1)、[Wiki-项目配置说明](https://github.com/binary-husky/gpt_academic/wiki/项目配置说明)。 - 「 程序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。如您能理解该读取逻辑，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（仅复制您修改过的配置条目即可）。 」 + 「 程序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。如您能理解以上读取逻辑，我们强烈建议您在`config.py`同路径下创建一个名为`config_private.py`的新配置文件，并使用`config_private.py`配置项目，以确保更新或其他用户无法轻易查看您的私有配置 」。 - 「 支持通过`环境变量`配置项目，环境变量的书写格式参考`docker-compose.yml`文件或者我们的[Wiki页面](https://github.com/binary-husky/gpt_academic/wiki/项目配置说明)。配置读取优先级: `环境变量` > `config_private.py` > `config.py`。 」 + 「 支持通过`环境变量`配置项目，环境变量的书写格式参考`docker-compose.yml`文件或者我们的[Wiki页面](https://github.com/binary-husky/gpt_academic/wiki/项目配置说明)。配置读取优先级: `环境变量` > `config_private.py` > `config.py` 」。 3. 安装依赖 @@ -149,6 +179,14 @@ git clone --depth=1 https://github.com/OpenLMLab/MOSS.git request_llms/moss # # 【可选步骤IV】确保config.py配置文件的AVAIL_LLM_MODELS包含了期望的模型，目前支持的全部模型如下(jittorllms系列目前仅支持docker方案)： AVAIL_LLM_MODELS = ["gpt-3.5-turbo", "api2d-gpt-3.5-turbo", "gpt-4", "api2d-gpt-4", "chatglm", "moss"] # + ["jittorllms_rwkv", "jittorllms_pangualpha", "jittorllms_llama"] + +# 【可选步骤V】支持本地模型INT8,INT4量化（这里所指的模型本身不是量化版本，目前deepseek-coder支持，后面测试后会加入更多模型量化选择） +pip install bitsandbyte +# windows用户安装bitsandbytes需要使用下面bitsandbytes-windows-webui +python -m pip install bitsandbytes --prefer-binary --extra-index-url=https://jllllll.github.io/bitsandbytes-windows-webui +pip install -U git+https://github.com/huggingface/transformers.git +pip install -U git+https://github.com/huggingface/accelerate.git +pip install peft ```

@@ -163,7 +201,7 @@ AVAIL_LLM_MODELS = ["gpt-3.5-turbo", "api2d-gpt-3.5-turbo", "gpt-4", "api2d-gpt- ### 安装方法II：使用Docker -0. 部署项目的全部能力（这个是包含cuda和latex的大型镜像。但如果您网速慢、硬盘小，则不推荐使用这个） +0. 部署项目的全部能力（这个是包含cuda和latex的大型镜像。但如果您网速慢、硬盘小，则不推荐该方法部署完整项目） [![fullcapacity](https://github.com/binary-husky/gpt_academic/actions/workflows/build-with-all-capacity.yml/badge.svg?branch=master)](https://github.com/binary-husky/gpt_academic/actions/workflows/build-with-all-capacity.yml) ``` sh @@ -192,26 +230,26 @@ P.S. 如果需要依赖Latex的插件功能，请见Wiki。另外，您也可以 ``` -### 安装方法III：其他部署姿势 +### 安装方法III：其他部署方法 1. **Windows一键运行脚本**。 -完全不熟悉python环境的Windows用户可以下载[Release](https://github.com/binary-husky/gpt_academic/releases)中发布的一键运行脚本安装无本地模型的版本。 -脚本的贡献来源是[oobabooga](https://github.com/oobabooga/one-click-installers)。 +完全不熟悉python环境的Windows用户可以下载[Release](https://github.com/binary-husky/gpt_academic/releases)中发布的一键运行脚本安装无本地模型的版本。脚本贡献来源：[oobabooga](https://github.com/oobabooga/one-click-installers)。 2. 使用第三方API、Azure等、文心一言、星火等，见[Wiki页面](https://github.com/binary-husky/gpt_academic/wiki/项目配置说明) 3. 云服务器远程部署避坑指南。 请访问[云服务器远程部署wiki](https://github.com/binary-husky/gpt_academic/wiki/%E4%BA%91%E6%9C%8D%E5%8A%A1%E5%99%A8%E8%BF%9C%E7%A8%8B%E9%83%A8%E7%BD%B2%E6%8C%87%E5%8D%97) -4. 一些新型的部署平台或方法 +4. 在其他平台部署&二级网址部署 - 使用Sealos[一键部署](https://github.com/binary-husky/gpt_academic/issues/993)。 - 使用WSL2（Windows Subsystem for Linux 子系统）。请访问[部署wiki-2](https://github.com/binary-husky/gpt_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) - 如何在二级网址（如`http://localhost/subpath`）下运行。请访问[FastAPI运行说明](docs/WithFastapi.md) +

# Advanced Usage ### I：自定义新的便捷按钮（学术快捷键） -任意文本编辑器打开`core_functional.py`，添加条目如下，然后重启程序。（如按钮已存在，那么前缀、后缀都支持热修改，无需重启程序即可生效。） +任意文本编辑器打开`core_functional.py`，添加如下条目，然后重启程序。（如果按钮已存在，那么可以直接修改（前缀、后缀都已支持热修改），无需重启程序即可生效。） 例如 ```python @@ -233,6 +271,7 @@ P.S. 如果需要依赖Latex的插件功能，请见Wiki。另外，您也可以 本项目的插件编写、调试难度很低，只要您具备一定的python基础知识，就可以仿照我们提供的模板实现自己的插件功能。 详情请参考[函数插件指南](https://github.com/binary-husky/gpt_academic/wiki/%E5%87%BD%E6%95%B0%E6%8F%92%E4%BB%B6%E6%8C%87%E5%8D%97)。 +

# Updates ### I：动态 @@ -332,7 +371,7 @@ GPT Academic开发者QQ群：`610599535` - 已知问题 - 某些浏览器翻译插件干扰此软件前端的运行 - - 官方Gradio目前有很多兼容性Bug，请务必使用`requirement.txt`安装Gradio + - 官方Gradio目前有很多兼容性问题，请**务必使用`requirement.txt`安装Gradio** ### III：主题 可以通过修改`THEME`选项（config.py）变更主题 @@ -343,8 +382,8 @@ GPT Academic开发者QQ群：`610599535` 1. `master` 分支: 主分支，稳定版 2. `frontier` 分支: 开发分支，测试版 -3. 如何接入其他大模型：[接入其他大模型](request_llms/README.md) - +3. 如何[接入其他大模型](request_llms/README.md) +4. 访问GPT-Academic的[在线服务并支持我们](https://github.com/binary-husky/gpt_academic/wiki/online) ### V：参考与学习 diff --git a/app.py b/app.py index 841ab9a0..4a15e07e 100644 --- a/app.py +++ b/app.py @@ -1,6 +1,17 @@ import os; os.environ['no_proxy'] = '*' # 避免代理网络产生意外污染 -import pickle -import base64 + +help_menu_description = \ +"""Github源代码开源和更新[地址🚀](https://github.com/binary-husky/gpt_academic), +感谢热情的[开发者们❤️](https://github.com/binary-husky/gpt_academic/graphs/contributors). +

常见问题请查阅[项目Wiki](https://github.com/binary-husky/gpt_academic/wiki), +如遇到Bug请前往[Bug反馈](https://github.com/binary-husky/gpt_academic/issues). +

普通对话使用说明: 1. 输入问题; 2. 点击提交 +

基础功能区使用说明: 1. 输入文本; 2. 点击任意基础功能区按钮 +

函数插件区使用说明: 1. 输入路径/问题, 或者上传文件; 2. 点击任意函数插件区按钮 +

虚空终端使用说明: 点击虚空终端, 然后根据提示输入指令, 再次点击虚空终端 +

如何保存对话: 点击保存当前的对话按钮 +

如何语音对话: 请阅读Wiki +

如何临时更换API_KEY: 在输入区输入临时API_KEY后提交（网页刷新后失效）""" def main(): import subprocess, sys @@ -10,7 +21,7 @@ def main(): raise ModuleNotFoundError("使用项目内置Gradio获取最优体验! 请运行 `pip install -r requirements.txt` 指令安装内置Gradio及其他依赖, 详情信息见requirements.txt.") from request_llms.bridge_all import predict from toolbox import format_io, find_free_port, on_file_uploaded, on_report_generated, get_conf, ArgsGeneralWrapper, load_chat_cookies, DummyWith - # 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 + # 建议您复制一个config_private.py放自己的秘密, 如API和代理网址 proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION = get_conf('proxies', 'WEB_PORT', 'LLM_MODEL', 'CONCURRENT_COUNT', 'AUTHENTICATION') CHATBOT_HEIGHT, LAYOUT, AVAIL_LLM_MODELS, AUTO_CLEAR_TXT = get_conf('CHATBOT_HEIGHT', 'LAYOUT', 'AVAIL_LLM_MODELS', 'AUTO_CLEAR_TXT') ENABLE_AUDIO, AUTO_CLEAR_TXT, PATH_LOGGING, AVAIL_THEMES, THEME = get_conf('ENABLE_AUDIO', 'AUTO_CLEAR_TXT', 'PATH_LOGGING', 'AVAIL_THEMES', 'THEME') @@ -20,21 +31,11 @@ def main(): # 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 from check_proxy import get_current_version - from themes.theme import adjust_theme, advanced_css, theme_declaration, load_dynamic_theme - + from themes.theme import adjust_theme, advanced_css, theme_declaration + from themes.theme import js_code_for_css_changing, js_code_for_darkmode_init, js_code_for_toggle_darkmode, js_code_for_persistent_cookie_init + from themes.theme import load_dynamic_theme, to_cookie_str, from_cookie_str, init_cookie title_html = f"

GPT 学术优化 {get_current_version()}

{theme_declaration}" - description = "Github源代码开源和更新[地址🚀](https://github.com/binary-husky/gpt_academic), " - description += "感谢热情的[开发者们❤️](https://github.com/binary-husky/gpt_academic/graphs/contributors)." - description += "

常见问题请查阅[项目Wiki](https://github.com/binary-husky/gpt_academic/wiki), " - description += "如遇到Bug请前往[Bug反馈](https://github.com/binary-husky/gpt_academic/issues)." - description += "

普通对话使用说明: 1. 输入问题; 2. 点击提交" - description += "

基础功能区使用说明: 1. 输入文本; 2. 点击任意基础功能区按钮" - description += "

函数插件区使用说明: 1. 输入路径/问题, 或者上传文件; 2. 点击任意函数插件区按钮" - description += "

虚空终端使用说明: 点击虚空终端, 然后根据提示输入指令, 再次点击虚空终端" - description += "

如何保存对话: 点击保存当前的对话按钮" - description += "

如何语音对话: 请阅读Wiki" - description += "

如何临时更换API_KEY: 在输入区输入临时API_KEY后提交（网页刷新后失效）" - + # 问询记录, python 版本建议3.9+（越新越好） import logging, uuid os.makedirs(PATH_LOGGING, exist_ok=True) @@ -88,7 +89,7 @@ def main(): with gr_L2(scale=1, elem_id="gpt-panel"): with gr.Accordion("输入区", open=True, elem_id="input-panel") as area_input_primary: with gr.Row(): - txt = gr.Textbox(show_label=False, lines=2, placeholder="输入问题或API密钥，输入多个密钥时，用英文逗号间隔。支持OpenAI密钥和API2D密钥共存。").style(container=False) + txt = gr.Textbox(show_label=False, lines=2, placeholder="输入问题或API密钥，输入多个密钥时，用英文逗号间隔。支持多个OpenAI密钥共存。").style(container=False) with gr.Row(): submitBtn = gr.Button("提交", elem_id="elem_submit", variant="primary") with gr.Row(): @@ -149,7 +150,7 @@ def main(): with gr.Row(): with gr.Tab("上传文件", elem_id="interact-panel"): gr.Markdown("请上传本地文件/压缩包供“函数插件区”功能调用。请注意: 上传文件后会自动把输入区修改为相应路径。") - file_upload_2 = gr.Files(label="任何文件, 推荐上传压缩文件(zip, tar)", file_count="multiple") + file_upload_2 = gr.Files(label="任何文件, 推荐上传压缩文件(zip, tar)", file_count="multiple", elem_id="elem_upload_float") with gr.Tab("更换模型 & Prompt", elem_id="interact-panel"): md_dropdown = gr.Dropdown(AVAIL_LLM_MODELS, value=LLM_MODEL, label="更换LLM模型/请求源").style(container=False) @@ -165,39 +166,24 @@ def main(): checkboxes_2 = gr.CheckboxGroup(["自定义菜单"], value=[], label="显示/隐藏自定义菜单", elem_id='cbs').style(container=False) dark_mode_btn = gr.Button("切换界面明暗 ☀", variant="secondary").style(size="sm") - dark_mode_btn.click(None, None, None, _js="""() => { - if (document.querySelectorAll('.dark').length) { - document.querySelectorAll('.dark').forEach(el => el.classList.remove('dark')); - } else { - document.querySelector('body').classList.add('dark'); - } - }""", + dark_mode_btn.click(None, None, None, _js=js_code_for_toggle_darkmode, ) with gr.Tab("帮助", elem_id="interact-panel"): - gr.Markdown(description) + gr.Markdown(help_menu_description) with gr.Floating(init_x="20%", init_y="50%", visible=False, width="40%", drag="top") as area_input_secondary: with gr.Accordion("浮动输入区", open=True, elem_id="input-panel2"): with gr.Row() as row: row.style(equal_height=True) with gr.Column(scale=10): - txt2 = gr.Textbox(show_label=False, placeholder="Input question here.", lines=8, label="输入区2").style(container=False) + txt2 = gr.Textbox(show_label=False, placeholder="Input question here.", + elem_id='user_input_float', lines=8, label="输入区2").style(container=False) with gr.Column(scale=1, min_width=40): submitBtn2 = gr.Button("提交", variant="primary"); submitBtn2.style(size="sm") resetBtn2 = gr.Button("重置", variant="secondary"); resetBtn2.style(size="sm") stopBtn2 = gr.Button("停止", variant="secondary"); stopBtn2.style(size="sm") clearBtn2 = gr.Button("清除", variant="secondary", visible=False); clearBtn2.style(size="sm") - def to_cookie_str(d): - # Pickle the dictionary and encode it as a string - pickled_dict = pickle.dumps(d) - cookie_value = base64.b64encode(pickled_dict).decode('utf-8') - return cookie_value - - def from_cookie_str(c): - # Decode the base64-encoded string and unpickle it into a dictionary - pickled_dict = base64.b64decode(c.encode('utf-8')) - return pickle.loads(pickled_dict) with gr.Floating(init_x="20%", init_y="50%", visible=False, width="40%", drag="top") as area_customize: with gr.Accordion("自定义菜单", open=True, elem_id="edit-panel"): @@ -229,11 +215,11 @@ def main(): else: ret.update({predefined_btns[basic_btn_dropdown_]: gr.update(visible=True, value=basic_fn_title)}) ret.update({cookies: cookies_}) - try: persistent_cookie_ = from_cookie_str(persistent_cookie_) # persistent cookie to dict + try: persistent_cookie_ = from_cookie_str(persistent_cookie_) # persistent cookie to dict except: persistent_cookie_ = {} - persistent_cookie_["custom_bnt"] = customize_fn_overwrite_ # dict update new value - persistent_cookie_ = to_cookie_str(persistent_cookie_) # persistent cookie to dict - ret.update({persistent_cookie: persistent_cookie_}) # write persistent cookie + persistent_cookie_["custom_bnt"] = customize_fn_overwrite_ # dict update new value + persistent_cookie_ = to_cookie_str(persistent_cookie_) # persistent cookie to dict + ret.update({persistent_cookie: persistent_cookie_}) # write persistent cookie return ret def reflesh_btn(persistent_cookie_, cookies_): @@ -254,10 +240,11 @@ def main(): else: ret.update({predefined_btns[k]: gr.update(visible=True, value=v['Title'])}) return ret - basic_fn_load.click(reflesh_btn, [persistent_cookie, cookies],[cookies, *customize_btns.values(), *predefined_btns.values()]) + basic_fn_load.click(reflesh_btn, [persistent_cookie, cookies], [cookies, *customize_btns.values(), *predefined_btns.values()]) h = basic_fn_confirm.click(assign_btn, [persistent_cookie, cookies, basic_btn_dropdown, basic_fn_title, basic_fn_prefix, basic_fn_suffix], [persistent_cookie, cookies, *customize_btns.values(), *predefined_btns.values()]) - h.then(None, [persistent_cookie], None, _js="""(persistent_cookie)=>{setCookie("persistent_cookie", persistent_cookie, 5);}""") # save persistent cookie + # save persistent cookie + h.then(None, [persistent_cookie], None, _js="""(persistent_cookie)=>{setCookie("persistent_cookie", persistent_cookie, 5);}""") # 功能区显示开关与功能区的互动 def fn_area_visibility(a): @@ -307,8 +294,8 @@ def main(): click_handle = btn.click(fn=ArgsGeneralWrapper(predict), inputs=[*input_combo, gr.State(True), gr.State(btn.value)], 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, txt2, checkboxes, cookies], [chatbot, txt, txt2, cookies]) - file_upload_2.upload(on_file_uploaded, [file_upload_2, chatbot, txt, txt2, checkboxes, cookies], [chatbot, txt, txt2, cookies]) + file_upload.upload(on_file_uploaded, [file_upload, chatbot, txt, txt2, checkboxes, cookies], [chatbot, txt, txt2, cookies]).then(None, None, None, _js=r"()=>{toast_push('上传完毕 ...'); cancel_loading_status();}") + file_upload_2.upload(on_file_uploaded, [file_upload_2, chatbot, txt, txt2, checkboxes, cookies], [chatbot, txt, txt2, cookies]).then(None, None, None, _js=r"()=>{toast_push('上传完毕 ...'); cancel_loading_status();}") # 函数插件-固定按钮区 for k in plugins: if not plugins[k].get("AsButton", True): continue @@ -344,18 +331,7 @@ def main(): None, [secret_css], None, - _js="""(css) => { - var existingStyles = document.querySelectorAll("style[data-loaded-css]"); - for (var i = 0; i < existingStyles.length; i++) { - var style = existingStyles[i]; - style.parentNode.removeChild(style); - } - var styleElement = document.createElement('style'); - styleElement.setAttribute('data-loaded-css', css); - styleElement.innerHTML = css; - document.head.appendChild(styleElement); - } - """ + _js=js_code_for_css_changing ) # 随变按钮的回调函数注册 def route(request: gr.Request, k, *args, **kwargs): @@ -387,27 +363,10 @@ def main(): rad.feed(cookies['uuid'].hex, audio) audio_mic.stream(deal_audio, inputs=[audio_mic, cookies]) - def init_cookie(cookies, chatbot): - # 为每一位访问的用户赋予一个独一无二的uuid编码 - cookies.update({'uuid': uuid.uuid4()}) - return cookies + demo.load(init_cookie, inputs=[cookies, chatbot], outputs=[cookies]) - darkmode_js = """(dark) => { - dark = dark == "True"; - if (document.querySelectorAll('.dark').length) { - if (!dark){ - document.querySelectorAll('.dark').forEach(el => el.classList.remove('dark')); - } - } else { - if (dark){ - document.querySelector('body').classList.add('dark'); - } - } - }""" - load_cookie_js = """(persistent_cookie) => { - return getCookie("persistent_cookie"); - }""" - demo.load(None, inputs=None, outputs=[persistent_cookie], _js=load_cookie_js) + darkmode_js = js_code_for_darkmode_init + demo.load(None, inputs=None, outputs=[persistent_cookie], _js=js_code_for_persistent_cookie_init) demo.load(None, inputs=[dark_mode], outputs=None, _js=darkmode_js) # 配置暗色主题或亮色主题 demo.load(None, inputs=[gr.Textbox(LAYOUT, visible=False)], outputs=None, _js='(LAYOUT)=>{GptAcademicJavaScriptInit(LAYOUT);}') @@ -418,8 +377,18 @@ def main(): if DARK_MODE: print(f"\t「暗色主题已启用（支持动态切换主题）」: http://localhost:{PORT}") else: print(f"\t「亮色主题已启用（支持动态切换主题）」: http://localhost:{PORT}") + def auto_updates(): time.sleep(0); auto_update() + def open_browser(): time.sleep(2); webbrowser.open_new_tab(f"http://localhost:{PORT}") + def warm_up_mods(): time.sleep(6); warm_up_modules() + + threading.Thread(target=auto_updates, name="self-upgrade", daemon=True).start() # 查看自动更新 + threading.Thread(target=open_browser, name="open-browser", daemon=True).start() # 打开浏览器页面 + threading.Thread(target=warm_up_mods, name="warm-up", daemon=True).start() # 预热tiktoken模块 + + run_delayed_tasks() demo.queue(concurrency_count=CONCURRENT_COUNT).launch(server_name="0.0.0.0", share=False, favicon_path="docs/logo.png", blocked_paths=["config.py","config_private.py","docker-compose.yml","Dockerfile"]) + # 如果需要在二级路径下运行 # CUSTOM_PATH = get_conf('CUSTOM_PATH') # if CUSTOM_PATH != "/": diff --git a/check_proxy.py b/check_proxy.py index e2ba3f1a..2df81855 100644 --- a/check_proxy.py +++ b/check_proxy.py @@ -159,7 +159,15 @@ def warm_up_modules(): enc.encode("模块预热", disallowed_special=()) enc = model_info["gpt-4"]['tokenizer'] enc.encode("模块预热", disallowed_special=()) + +def warm_up_vectordb(): + print('正在执行一些模块的预热 ...') + from toolbox import ProxyNetworkActivate + with ProxyNetworkActivate("Warmup_Modules"): + import nltk + with ProxyNetworkActivate("Warmup_Modules"): nltk.download("punkt") + if __name__ == '__main__': import os os.environ['no_proxy'] = '*' # 避免代理网络产生意外污染 diff --git a/config.py b/config.py index 46d1d90d..70fcaf5e 100644 --- a/config.py +++ b/config.py @@ -19,13 +19,13 @@ API_KEY = "此处填API密钥" # 可同时填写多个API-KEY，用英文逗 USE_PROXY = False if USE_PROXY: """ + 代理网络的地址，打开你的代理软件查看代理协议(socks5h / http)、地址(localhost)和端口(11284) 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 <配置教程&视频教程> https://github.com/binary-husky/gpt_academic/issues/1> [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） + [地址] 填localhost或者127.0.0.1（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*学*网软件查看代理的协议(socks5h / http)、地址(localhost)和端口(11284) proxies = { # [协议]:// [地址] :[端口] "http": "socks5h://localhost:11284", # 再例如 "http": "http://127.0.0.1:7890", @@ -99,14 +99,25 @@ AVAIL_LLM_MODELS = ["gpt-3.5-turbo-1106","gpt-4-1106-preview","gpt-4-vision-prev "api2d-gpt-3.5-turbo", 'api2d-gpt-3.5-turbo-16k', "gpt-4", "gpt-4-32k", "azure-gpt-4", "api2d-gpt-4", "chatglm3", "moss", "claude-2"] -# P.S. 其他可用的模型还包括 ["zhipuai", "qianfan", "deepseekcoder", "llama2", "qwen", "gpt-3.5-turbo-0613", "gpt-3.5-turbo-16k-0613", "gpt-3.5-random" -# "spark", "sparkv2", "sparkv3", "chatglm_onnx", "claude-1-100k", "claude-2", "internlm", "jittorllms_pangualpha", "jittorllms_llama"] +# P.S. 其他可用的模型还包括 ["zhipuai", "qianfan", "deepseekcoder", "llama2", "qwen-local", "gpt-3.5-turbo-0613", "gpt-3.5-turbo-16k-0613", "gpt-3.5-random" +# "spark", "sparkv2", "sparkv3", "chatglm_onnx", "claude-1-100k", "claude-2", "internlm", "jittorllms_pangualpha", "jittorllms_llama" +# “qwen-turbo", "qwen-plus", "qwen-max"] # 定义界面上“询问多个GPT模型”插件应该使用哪些模型，请从AVAIL_LLM_MODELS中选择，并在不同模型之间用`&`间隔，例如"gpt-3.5-turbo&chatglm3&azure-gpt-4" MULTI_QUERY_LLM_MODELS = "gpt-3.5-turbo&chatglm3" +# 选择本地模型变体（只有当AVAIL_LLM_MODELS包含了对应本地模型时，才会起作用） +# 如果你选择Qwen系列的模型，那么请在下面的QWEN_MODEL_SELECTION中指定具体的模型 +# 也可以是具体的模型路径 +QWEN_LOCAL_MODEL_SELECTION = "Qwen/Qwen-1_8B-Chat-Int8" + + +# 接入通义千问在线大模型 https://dashscope.console.aliyun.com/ +DASHSCOPE_API_KEY = "" # 阿里灵积云API_KEY + + # 百度千帆（LLM_MODEL="qianfan"） BAIDU_CLOUD_API_KEY = '' BAIDU_CLOUD_SECRET_KEY = '' @@ -121,7 +132,6 @@ CHATGLM_PTUNING_CHECKPOINT = "" # 例如"/home/hmp/ChatGLM2-6B/ptuning/output/6b LOCAL_MODEL_DEVICE = "cpu" # 可选 "cuda" LOCAL_MODEL_QUANT = "FP16" # 默认 "FP16" "INT4" 启用量化INT4版本 "INT8" 启用量化INT8版本 - # 设置gradio的并行线程数（不需要修改） CONCURRENT_COUNT = 100 @@ -239,6 +249,10 @@ WHEN_TO_USE_PROXY = ["Download_LLM", "Download_Gradio_Theme", "Connect_Grobid", BLOCK_INVALID_APIKEY = False +# 启用插件热加载 +PLUGIN_HOT_RELOAD = False + + # 自定义按钮的最大数量限制 NUM_CUSTOM_BASIC_BTN = 4 @@ -282,6 +296,9 @@ NUM_CUSTOM_BASIC_BTN = 4 │ ├── ZHIPUAI_API_KEY │ └── ZHIPUAI_MODEL │ +├── "qwen-turbo" 等通义千问大模型 +│ └── DASHSCOPE_API_KEY +│ └── "newbing" Newbing接口不再稳定，不推荐使用 ├── NEWBING_STYLE └── NEWBING_COOKIES @@ -298,7 +315,7 @@ NUM_CUSTOM_BASIC_BTN = 4 ├── "jittorllms_pangualpha" ├── "jittorllms_llama" ├── "deepseekcoder" -├── "qwen" +├── "qwen-local" ├── RWKV的支持见Wiki └── "llama2" diff --git a/crazy_functional.py b/crazy_functional.py index 3d4df718..c323dc32 100644 --- a/crazy_functional.py +++ b/crazy_functional.py @@ -345,7 +345,7 @@ def get_crazy_functions(): "Color": "stop", "AsButton": False, "AdvancedArgs": True, # 调用时，唤起高级参数输入区（默认False） - "ArgsReminder": "支持任意数量的llm接口，用&符号分隔。例如chatglm&gpt-3.5-turbo&api2d-gpt-4", # 高级参数输入区的显示提示 + "ArgsReminder": "支持任意数量的llm接口，用&符号分隔。例如chatglm&gpt-3.5-turbo&gpt-4", # 高级参数输入区的显示提示 "Function": HotReload(同时问询_指定模型) }, }) @@ -354,9 +354,9 @@ def get_crazy_functions(): print('Load function plugin failed') try: - from crazy_functions.图片生成 import 图片生成_DALLE2, 图片生成_DALLE3 + from crazy_functions.图片生成 import 图片生成_DALLE2, 图片生成_DALLE3, 图片修改_DALLE2 function_plugins.update({ - "图片生成_DALLE2 （先切换模型到openai或api2d）": { + "图片生成_DALLE2 （先切换模型到gpt-*）": { "Group": "对话", "Color": "stop", "AsButton": False, @@ -367,16 +367,26 @@ def get_crazy_functions(): }, }) function_plugins.update({ - "图片生成_DALLE3 （先切换模型到openai或api2d）": { + "图片生成_DALLE3 （先切换模型到gpt-*）": { "Group": "对话", "Color": "stop", "AsButton": False, "AdvancedArgs": True, # 调用时，唤起高级参数输入区（默认False） - "ArgsReminder": "在这里输入分辨率, 如1024x1024（默认），支持 1024x1024, 1792x1024, 1024x1792。如需生成高清图像，请输入 1024x1024-HD, 1792x1024-HD, 1024x1792-HD。", # 高级参数输入区的显示提示 + "ArgsReminder": "在这里输入自定义参数「分辨率-质量(可选)-风格(可选)」, 参数示例「1024x1024-hd-vivid」 || 分辨率支持 「1024x1024」(默认) /「1792x1024」/「1024x1792」 || 质量支持 「-standard」(默认) /「-hd」 || 风格支持 「-vivid」(默认) /「-natural」", # 高级参数输入区的显示提示 "Info": "使用DALLE3生成图片 | 输入参数字符串，提供图像的内容", "Function": HotReload(图片生成_DALLE3) }, }) + function_plugins.update({ + "图片修改_DALLE2 （先切换模型到gpt-*）": { + "Group": "对话", + "Color": "stop", + "AsButton": False, + "AdvancedArgs": False, # 调用时，唤起高级参数输入区（默认False） + # "Info": "使用DALLE2修改图片 | 输入参数字符串，提供图像的内容", + "Function": HotReload(图片修改_DALLE2) + }, + }) except: print(trimmed_format_exc()) print('Load function plugin failed') @@ -430,7 +440,7 @@ def get_crazy_functions(): print('Load function plugin failed') try: - from crazy_functions.Langchain知识库 import 知识库问答 + from crazy_functions.知识库问答 import 知识库文件注入 function_plugins.update({ "构建知识库（先上传文件素材,再运行此插件）": { "Group": "对话", @@ -438,7 +448,7 @@ def get_crazy_functions(): "AsButton": False, "AdvancedArgs": True, "ArgsReminder": "此处待注入的知识库名称id, 默认为default。文件进入知识库后可长期保存。可以通过再次调用本插件的方式，向知识库追加更多文档。", - "Function": HotReload(知识库问答) + "Function": HotReload(知识库文件注入) } }) except: @@ -446,9 +456,9 @@ def get_crazy_functions(): print('Load function plugin failed') try: - from crazy_functions.Langchain知识库 import 读取知识库作答 + from crazy_functions.知识库问答 import 读取知识库作答 function_plugins.update({ - "知识库问答（构建知识库后,再运行此插件）": { + "知识库文件注入（构建知识库后,再运行此插件）": { "Group": "对话", "Color": "stop", "AsButton": False, @@ -489,7 +499,7 @@ def get_crazy_functions(): }) from crazy_functions.Latex输出PDF结果 import Latex翻译中文并重新编译PDF function_plugins.update({ - "Arixv论文精细翻译（输入arxivID）[需Latex]": { + "Arxiv论文精细翻译（输入arxivID）[需Latex]": { "Group": "学术", "Color": "stop", "AsButton": False, @@ -580,6 +590,20 @@ def get_crazy_functions(): print(trimmed_format_exc()) print('Load function plugin failed') + try: + from crazy_functions.互动小游戏 import 随机小游戏 + function_plugins.update({ + "随机互动小游戏（仅供测试）": { + "Group": "智能体", + "Color": "stop", + "AsButton": False, + "Function": HotReload(随机小游戏) + } + }) + except: + print(trimmed_format_exc()) + print('Load function plugin failed') + # try: # from crazy_functions.chatglm微调工具 import 微调数据集生成 # function_plugins.update({ diff --git a/crazy_functions/Latex全文润色.py b/crazy_functions/Latex全文润色.py index 0bc7d401..b736fe89 100644 --- a/crazy_functions/Latex全文润色.py +++ b/crazy_functions/Latex全文润色.py @@ -26,8 +26,8 @@ class PaperFileGroup(): 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) + from crazy_functions.pdf_fns.breakdown_txt import breakdown_text_to_satisfy_token_limit + segments = breakdown_text_to_satisfy_token_limit(file_content, max_token_limit) for j, segment in enumerate(segments): self.sp_file_contents.append(segment) self.sp_file_index.append(index) diff --git a/crazy_functions/Latex全文翻译.py b/crazy_functions/Latex全文翻译.py index 846bd80d..49470c86 100644 --- a/crazy_functions/Latex全文翻译.py +++ b/crazy_functions/Latex全文翻译.py @@ -26,8 +26,8 @@ class PaperFileGroup(): 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) + from crazy_functions.pdf_fns.breakdown_txt import breakdown_text_to_satisfy_token_limit + segments = breakdown_text_to_satisfy_token_limit(file_content, max_token_limit) for j, segment in enumerate(segments): self.sp_file_contents.append(segment) self.sp_file_index.append(index) diff --git a/crazy_functions/Latex输出PDF结果.py b/crazy_functions/Latex输出PDF结果.py index f3919edc..18a8d1ba 100644 --- a/crazy_functions/Latex输出PDF结果.py +++ b/crazy_functions/Latex输出PDF结果.py @@ -88,6 +88,9 @@ def arxiv_download(chatbot, history, txt, allow_cache=True): target_file = pj(translation_dir, 'translate_zh.pdf') if os.path.exists(target_file): promote_file_to_downloadzone(target_file, rename_file=None, chatbot=chatbot) + target_file_compare = pj(translation_dir, 'comparison.pdf') + if os.path.exists(target_file_compare): + promote_file_to_downloadzone(target_file_compare, rename_file=None, chatbot=chatbot) return target_file return False def is_float(s): diff --git a/crazy_functions/crazy_utils.py b/crazy_functions/crazy_utils.py index afe079f4..4d3b1953 100644 --- a/crazy_functions/crazy_utils.py +++ b/crazy_functions/crazy_utils.py @@ -1,4 +1,4 @@ -from toolbox import update_ui, get_conf, trimmed_format_exc, get_max_token +from toolbox import update_ui, get_conf, trimmed_format_exc, get_max_token, Singleton import threading import os import logging @@ -139,6 +139,8 @@ def can_multi_process(llm): if llm.startswith('gpt-'): return True if llm.startswith('api2d-'): return True if llm.startswith('azure-'): return True + if llm.startswith('spark'): return True + if llm.startswith('zhipuai'): return True return False def request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency( @@ -312,95 +314,6 @@ def request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency( return gpt_response_collection -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: - 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 force_breakdown(txt, limit, get_token_fn): - """ - 当无法用标点、空行分割时，我们用最暴力的方法切割 - """ - for i in reversed(range(len(txt))): - if get_token_fn(txt[:i]) < limit: - return txt[:i], txt[i:] - return "Tiktoken未知错误", "Tiktoken未知错误" - -def breakdown_txt_to_satisfy_token_limit_for_pdf(txt, get_token_fn, limit): - # 递归 - def cut(txt_tocut, must_break_at_empty_line, break_anyway=False): - 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 - prev = "\n".join(lines[:cnt]) - post = "\n".join(lines[cnt:]) - if get_token_fn(prev) < limit: - break - if cnt == 0: - if break_anyway: - prev, post = force_breakdown(txt_tocut, limit, get_token_fn) - else: - raise RuntimeError(f"存在一行极长的文本！{txt_tocut}") - # print(len(post)) - # 列表递归接龙 - result = [prev] - result.extend(cut(post, must_break_at_empty_line, break_anyway=break_anyway)) - return result - try: - # 第1次尝试，将双空行（\n\n）作为切分点 - return cut(txt, must_break_at_empty_line=True) - except RuntimeError: - try: - # 第2次尝试，将单空行（\n）作为切分点 - return cut(txt, must_break_at_empty_line=False) - except RuntimeError: - try: - # 第3次尝试，将英文句号（.）作为切分点 - res = cut(txt.replace('.', '。\n'), must_break_at_empty_line=False) # 这个中文的句号是故意的，作为一个标识而存在 - return [r.replace('。\n', '.') for r in res] - except RuntimeError as e: - try: - # 第4次尝试，将中文句号（。）作为切分点 - res = cut(txt.replace('。', '。。\n'), must_break_at_empty_line=False) - return [r.replace('。。\n', '。') for r in res] - except RuntimeError as e: - # 第5次尝试，没办法了，随便切一下敷衍吧 - return cut(txt, must_break_at_empty_line=False, break_anyway=True) - - def read_and_clean_pdf_text(fp): """ @@ -631,90 +544,6 @@ def get_files_from_everything(txt, type): # type='.md' - -def Singleton(cls): - _instance = {} - - def _singleton(*args, **kargs): - if cls not in _instance: - _instance[cls] = cls(*args, **kargs) - return _instance[cls] - - return _singleton - - -@Singleton -class knowledge_archive_interface(): - def __init__(self) -> None: - self.threadLock = threading.Lock() - self.current_id = "" - self.kai_path = None - self.qa_handle = None - self.text2vec_large_chinese = None - - def get_chinese_text2vec(self): - if self.text2vec_large_chinese is None: - # < -------------------预热文本向量化模组--------------- > - from toolbox import ProxyNetworkActivate - print('Checking Text2vec ...') - from langchain.embeddings.huggingface import HuggingFaceEmbeddings - with ProxyNetworkActivate('Download_LLM'): # 临时地激活代理网络 - self.text2vec_large_chinese = HuggingFaceEmbeddings(model_name="GanymedeNil/text2vec-large-chinese") - - return self.text2vec_large_chinese - - - def feed_archive(self, file_manifest, id="default"): - self.threadLock.acquire() - # import uuid - self.current_id = id - from zh_langchain import construct_vector_store - self.qa_handle, self.kai_path = construct_vector_store( - vs_id=self.current_id, - files=file_manifest, - sentence_size=100, - history=[], - one_conent="", - one_content_segmentation="", - text2vec = self.get_chinese_text2vec(), - ) - self.threadLock.release() - - def get_current_archive_id(self): - return self.current_id - - def get_loaded_file(self): - return self.qa_handle.get_loaded_file() - - def answer_with_archive_by_id(self, txt, id): - self.threadLock.acquire() - if not self.current_id == id: - self.current_id = id - from zh_langchain import construct_vector_store - self.qa_handle, self.kai_path = construct_vector_store( - vs_id=self.current_id, - files=[], - sentence_size=100, - history=[], - one_conent="", - one_content_segmentation="", - text2vec = self.get_chinese_text2vec(), - ) - VECTOR_SEARCH_SCORE_THRESHOLD = 0 - VECTOR_SEARCH_TOP_K = 4 - CHUNK_SIZE = 512 - resp, prompt = self.qa_handle.get_knowledge_based_conent_test( - query = txt, - vs_path = self.kai_path, - score_threshold=VECTOR_SEARCH_SCORE_THRESHOLD, - vector_search_top_k=VECTOR_SEARCH_TOP_K, - chunk_conent=True, - chunk_size=CHUNK_SIZE, - text2vec = self.get_chinese_text2vec(), - ) - self.threadLock.release() - return resp, prompt - @Singleton class nougat_interface(): def __init__(self): diff --git a/crazy_functions/game_fns/game_ascii_art.py b/crazy_functions/game_fns/game_ascii_art.py new file mode 100644 index 00000000..e0b70087 --- /dev/null +++ b/crazy_functions/game_fns/game_ascii_art.py @@ -0,0 +1,42 @@ +from toolbox import CatchException, update_ui, update_ui_lastest_msg +from crazy_functions.multi_stage.multi_stage_utils import GptAcademicGameBaseState +from crazy_functions.crazy_utils import request_gpt_model_in_new_thread_with_ui_alive +from request_llms.bridge_all import predict_no_ui_long_connection +from crazy_functions.game_fns.game_utils import get_code_block, is_same_thing +import random + + +class MiniGame_ASCII_Art(GptAcademicGameBaseState): + def step(self, prompt, chatbot, history): + if self.step_cnt == 0: + chatbot.append(["我画你猜（动物）", "请稍等..."]) + else: + if prompt.strip() == 'exit': + self.delete_game = True + yield from update_ui_lastest_msg(lastmsg=f"谜底是{self.obj}，游戏结束。", chatbot=chatbot, history=history, delay=0.) + return + chatbot.append([prompt, ""]) + yield from update_ui(chatbot=chatbot, history=history) + + if self.step_cnt == 0: + self.lock_plugin(chatbot) + self.cur_task = 'draw' + + if self.cur_task == 'draw': + avail_obj = ["狗","猫","鸟","鱼","老鼠","蛇"] + self.obj = random.choice(avail_obj) + inputs = "I want to play a game called Guess the ASCII art. You can draw the ASCII art and I will try to guess it. " + \ + f"This time you draw a {self.obj}. Note that you must not indicate what you have draw in the text, and you should only produce the ASCII art wrapped by ```. " + raw_res = predict_no_ui_long_connection(inputs=inputs, llm_kwargs=self.llm_kwargs, history=[], sys_prompt="") + self.cur_task = 'identify user guess' + res = get_code_block(raw_res) + history += ['', f'the answer is {self.obj}', inputs, res] + yield from update_ui_lastest_msg(lastmsg=res, chatbot=chatbot, history=history, delay=0.) + + elif self.cur_task == 'identify user guess': + if is_same_thing(self.obj, prompt, self.llm_kwargs): + self.delete_game = True + yield from update_ui_lastest_msg(lastmsg="你猜对了！", chatbot=chatbot, history=history, delay=0.) + else: + self.cur_task = 'identify user guess' + yield from update_ui_lastest_msg(lastmsg="猜错了，再试试，输入“exit”获取答案。", chatbot=chatbot, history=history, delay=0.) \ No newline at end of file diff --git a/crazy_functions/game_fns/game_interactive_story.py b/crazy_functions/game_fns/game_interactive_story.py new file mode 100644 index 00000000..5c25f4a3 --- /dev/null +++ b/crazy_functions/game_fns/game_interactive_story.py @@ -0,0 +1,212 @@ +prompts_hs = """ 请以“{headstart}”为开头，编写一个小说的第一幕。 + +- 尽量短，不要包含太多情节，因为你接下来将会与用户互动续写下面的情节，要留出足够的互动空间。 +- 出现人物时，给出人物的名字。 +- 积极地运用环境描写、人物描写等手法，让读者能够感受到你的故事世界。 +- 积极地运用修辞手法，比如比喻、拟人、排比、对偶、夸张等等。 +- 字数要求：第一幕的字数少于300字，且少于2个段落。 +""" + +prompts_interact = """ 小说的前文回顾： +「 +{previously_on_story} +」 + +你是一个作家，根据以上的情节，给出4种不同的后续剧情发展方向，每个发展方向都精明扼要地用一句话说明。稍后，我将在这4个选择中，挑选一种剧情发展。 + +输出格式例如： +1. 后续剧情发展1 +2. 后续剧情发展2 +3. 后续剧情发展3 +4. 后续剧情发展4 +""" + + +prompts_resume = """小说的前文回顾： +「 +{previously_on_story} +」 + +你是一个作家，我们正在互相讨论，确定后续剧情的发展。 +在以下的剧情发展中， +「 +{choice} +」 +我认为更合理的是：{user_choice}。 +请在前文的基础上（不要重复前文），围绕我选定的剧情情节，编写小说的下一幕。 + +- 禁止杜撰不符合我选择的剧情。 +- 尽量短，不要包含太多情节，因为你接下来将会与用户互动续写下面的情节，要留出足够的互动空间。 +- 不要重复前文。 +- 出现人物时，给出人物的名字。 +- 积极地运用环境描写、人物描写等手法，让读者能够感受到你的故事世界。 +- 积极地运用修辞手法，比如比喻、拟人、排比、对偶、夸张等等。 +- 小说的下一幕字数少于300字，且少于2个段落。 +""" + + +prompts_terminate = """小说的前文回顾： +「 +{previously_on_story} +」 + +你是一个作家，我们正在互相讨论，确定后续剧情的发展。 +现在，故事该结束了，我认为最合理的故事结局是：{user_choice}。 + +请在前文的基础上（不要重复前文），编写小说的最后一幕。 + +- 不要重复前文。 +- 出现人物时，给出人物的名字。 +- 积极地运用环境描写、人物描写等手法，让读者能够感受到你的故事世界。 +- 积极地运用修辞手法，比如比喻、拟人、排比、对偶、夸张等等。 +- 字数要求：最后一幕的字数少于1000字。 +""" + + +from toolbox import CatchException, update_ui, update_ui_lastest_msg +from crazy_functions.multi_stage.multi_stage_utils import GptAcademicGameBaseState +from crazy_functions.crazy_utils import request_gpt_model_in_new_thread_with_ui_alive +from request_llms.bridge_all import predict_no_ui_long_connection +from crazy_functions.game_fns.game_utils import get_code_block, is_same_thing +import random + + +class MiniGame_ResumeStory(GptAcademicGameBaseState): + story_headstart = [ + '先行者知道，他现在是全宇宙中唯一的一个人了。', + '深夜，一个年轻人穿过天安门广场向纪念堂走去。在二十二世纪编年史中，计算机把他的代号定为M102。', + '他知道，这最后一课要提前讲了。又一阵剧痛从肝部袭来，几乎使他晕厥过去。', + '在距地球五万光年的远方，在银河系的中心，一场延续了两万年的星际战争已接近尾声。那里的太空中渐渐隐现出一个方形区域，仿佛灿烂的群星的背景被剪出一个方口。', + '伊依一行三人乘坐一艘游艇在南太平洋上做吟诗航行，他们的目的地是南极，如果几天后能顺利到达那里，他们将钻出地壳去看诗云。', + '很多人生来就会莫名其妙地迷上一样东西，仿佛他的出生就是要和这东西约会似的，正是这样，圆圆迷上了肥皂泡。' + ] + + + def begin_game_step_0(self, prompt, chatbot, history): + # init game at step 0 + self.headstart = random.choice(self.story_headstart) + self.story = [] + chatbot.append(["互动写故事", f"这次的故事开头是：{self.headstart}"]) + self.sys_prompt_ = '你是一个想象力丰富的杰出作家。正在与你的朋友互动，一起写故事，因此你每次写的故事段落应少于300字（结局除外）。' + + + def generate_story_image(self, story_paragraph): + try: + from crazy_functions.图片生成 import gen_image + prompt_ = predict_no_ui_long_connection(inputs=story_paragraph, llm_kwargs=self.llm_kwargs, history=[], sys_prompt='你需要根据用户给出的小说段落，进行简短的环境描写。要求：80字以内。') + image_url, image_path = gen_image(self.llm_kwargs, prompt_, '512x512', model="dall-e-2", quality='standard', style='natural') + return f'
' + except: + return '' + + def step(self, prompt, chatbot, history): + + """ + 首先，处理游戏初始化等特殊情况 + """ + if self.step_cnt == 0: + self.begin_game_step_0(prompt, chatbot, history) + self.lock_plugin(chatbot) + self.cur_task = 'head_start' + else: + if prompt.strip() == 'exit' or prompt.strip() == '结束剧情': + # should we terminate game here? + self.delete_game = True + yield from update_ui_lastest_msg(lastmsg=f"游戏结束。", chatbot=chatbot, history=history, delay=0.) + return + if '剧情收尾' in prompt: + self.cur_task = 'story_terminate' + # # well, game resumes + # chatbot.append([prompt, ""]) + # update ui, don't keep the user waiting + yield from update_ui(chatbot=chatbot, history=history) + + + """ + 处理游戏的主体逻辑 + """ + if self.cur_task == 'head_start': + """ + 这是游戏的第一步 + """ + inputs_ = prompts_hs.format(headstart=self.headstart) + history_ = [] + story_paragraph = yield from request_gpt_model_in_new_thread_with_ui_alive( + inputs_, '故事开头', self.llm_kwargs, + chatbot, history_, self.sys_prompt_ + ) + self.story.append(story_paragraph) + # # 配图 + yield from update_ui_lastest_msg(lastmsg=story_paragraph + '
正在生成插图中 ...', chatbot=chatbot, history=history, delay=0.) + yield from update_ui_lastest_msg(lastmsg=story_paragraph + '
'+ self.generate_story_image(story_paragraph), chatbot=chatbot, history=history, delay=0.) + + # # 构建后续剧情引导 + previously_on_story = "" + for s in self.story: + previously_on_story += s + '\n' + inputs_ = prompts_interact.format(previously_on_story=previously_on_story) + history_ = [] + self.next_choices = yield from request_gpt_model_in_new_thread_with_ui_alive( + inputs_, '请在以下几种故事走向中，选择一种（当然，您也可以选择给出其他故事走向）：', self.llm_kwargs, + chatbot, + history_, + self.sys_prompt_ + ) + self.cur_task = 'user_choice' + + + elif self.cur_task == 'user_choice': + """ + 根据用户的提示，确定故事的下一步 + """ + if '请在以下几种故事走向中，选择一种' in chatbot[-1][0]: chatbot.pop(-1) + previously_on_story = "" + for s in self.story: + previously_on_story += s + '\n' + inputs_ = prompts_resume.format(previously_on_story=previously_on_story, choice=self.next_choices, user_choice=prompt) + history_ = [] + story_paragraph = yield from request_gpt_model_in_new_thread_with_ui_alive( + inputs_, f'下一段故事（您的选择是：{prompt}）。', self.llm_kwargs, + chatbot, history_, self.sys_prompt_ + ) + self.story.append(story_paragraph) + # # 配图 + yield from update_ui_lastest_msg(lastmsg=story_paragraph + '
正在生成插图中 ...', chatbot=chatbot, history=history, delay=0.) + yield from update_ui_lastest_msg(lastmsg=story_paragraph + '
'+ self.generate_story_image(story_paragraph), chatbot=chatbot, history=history, delay=0.) + + # # 构建后续剧情引导 + previously_on_story = "" + for s in self.story: + previously_on_story += s + '\n' + inputs_ = prompts_interact.format(previously_on_story=previously_on_story) + history_ = [] + self.next_choices = yield from request_gpt_model_in_new_thread_with_ui_alive( + inputs_, + '请在以下几种故事走向中，选择一种。当然，您也可以给出您心中的其他故事走向。另外，如果您希望剧情立即收尾，请输入剧情走向，并以“剧情收尾”四个字提示程序。', self.llm_kwargs, + chatbot, + history_, + self.sys_prompt_ + ) + self.cur_task = 'user_choice' + + + elif self.cur_task == 'story_terminate': + """ + 根据用户的提示，确定故事的结局 + """ + previously_on_story = "" + for s in self.story: + previously_on_story += s + '\n' + inputs_ = prompts_terminate.format(previously_on_story=previously_on_story, user_choice=prompt) + history_ = [] + story_paragraph = yield from request_gpt_model_in_new_thread_with_ui_alive( + inputs_, f'故事收尾（您的选择是：{prompt}）。', self.llm_kwargs, + chatbot, history_, self.sys_prompt_ + ) + # # 配图 + yield from update_ui_lastest_msg(lastmsg=story_paragraph + '
正在生成插图中 ...', chatbot=chatbot, history=history, delay=0.) + yield from update_ui_lastest_msg(lastmsg=story_paragraph + '
'+ self.generate_story_image(story_paragraph), chatbot=chatbot, history=history, delay=0.) + + # terminate game + self.delete_game = True + return diff --git a/crazy_functions/game_fns/game_utils.py b/crazy_functions/game_fns/game_utils.py new file mode 100644 index 00000000..09b6f7a9 --- /dev/null +++ b/crazy_functions/game_fns/game_utils.py @@ -0,0 +1,35 @@ + +from crazy_functions.json_fns.pydantic_io import GptJsonIO, JsonStringError +from request_llms.bridge_all import predict_no_ui_long_connection +def get_code_block(reply): + import re + pattern = r"```([\s\S]*?)```" # regex pattern to match code blocks + matches = re.findall(pattern, reply) # find all code blocks in text + if len(matches) == 1: + return "```" + matches[0] + "```" # code block + raise RuntimeError("GPT is not generating proper code.") + +def is_same_thing(a, b, llm_kwargs): + from pydantic import BaseModel, Field + class IsSameThing(BaseModel): + is_same_thing: bool = Field(description="determine whether two objects are same thing.", default=False) + + def run_gpt_fn(inputs, sys_prompt, history=[]): + return predict_no_ui_long_connection( + inputs=inputs, llm_kwargs=llm_kwargs, + history=history, sys_prompt=sys_prompt, observe_window=[] + ) + + gpt_json_io = GptJsonIO(IsSameThing) + inputs_01 = "Identity whether the user input and the target is the same thing: \n target object: {a} \n user input object: {b} \n\n\n".format(a=a, b=b) + inputs_01 += "\n\n\n Note that the user may describe the target object with a different language, e.g. cat and 猫 are the same thing." + analyze_res_cot_01 = run_gpt_fn(inputs_01, "", []) + + inputs_02 = inputs_01 + gpt_json_io.format_instructions + analyze_res = run_gpt_fn(inputs_02, "", [inputs_01, analyze_res_cot_01]) + + try: + res = gpt_json_io.generate_output_auto_repair(analyze_res, run_gpt_fn) + return res.is_same_thing + except JsonStringError as e: + return False \ No newline at end of file diff --git a/crazy_functions/ipc_fns/mp.py b/crazy_functions/ipc_fns/mp.py new file mode 100644 index 00000000..575d47cc --- /dev/null +++ b/crazy_functions/ipc_fns/mp.py @@ -0,0 +1,37 @@ +import platform +import pickle +import multiprocessing + +def run_in_subprocess_wrapper_func(v_args): + func, args, kwargs, return_dict, exception_dict = pickle.loads(v_args) + import sys + try: + result = func(*args, **kwargs) + return_dict['result'] = result + except Exception as e: + exc_info = sys.exc_info() + exception_dict['exception'] = exc_info + +def run_in_subprocess_with_timeout(func, timeout=60): + if platform.system() == 'Linux': + def wrapper(*args, **kwargs): + return_dict = multiprocessing.Manager().dict() + exception_dict = multiprocessing.Manager().dict() + v_args = pickle.dumps((func, args, kwargs, return_dict, exception_dict)) + process = multiprocessing.Process(target=run_in_subprocess_wrapper_func, args=(v_args,)) + process.start() + process.join(timeout) + if process.is_alive(): + process.terminate() + raise TimeoutError(f'功能单元{str(func)}未能在规定时间内完成任务') + process.close() + if 'exception' in exception_dict: + # ooops, the subprocess ran into an exception + exc_info = exception_dict['exception'] + raise exc_info[1].with_traceback(exc_info[2]) + if 'result' in return_dict.keys(): + # If the subprocess ran successfully, return the result + return return_dict['result'] + return wrapper + else: + return func \ No newline at end of file diff --git a/crazy_functions/latex_fns/latex_actions.py b/crazy_functions/latex_fns/latex_actions.py index 74e8757e..8772f5e1 100644 --- a/crazy_functions/latex_fns/latex_actions.py +++ b/crazy_functions/latex_fns/latex_actions.py @@ -175,7 +175,6 @@ class LatexPaperFileGroup(): self.sp_file_contents = [] self.sp_file_index = [] self.sp_file_tag = [] - # count_token from request_llms.bridge_all import model_info enc = model_info["gpt-3.5-turbo"]['tokenizer'] @@ -192,13 +191,12 @@ class LatexPaperFileGroup(): 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) + from crazy_functions.pdf_fns.breakdown_txt import breakdown_text_to_satisfy_token_limit + segments = breakdown_text_to_satisfy_token_limit(file_content, 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 merge_result(self): self.file_result = ["" for _ in range(len(self.file_paths))] @@ -404,7 +402,7 @@ def 编译Latex(chatbot, history, main_file_original, main_file_modified, work_f result_pdf = pj(work_folder_modified, f'merge_diff.pdf') # get pdf path promote_file_to_downloadzone(result_pdf, rename_file=None, chatbot=chatbot) # promote file to web UI if modified_pdf_success: - yield from update_ui_lastest_msg(f'转化PDF编译已经成功, 即将退出 ...', chatbot, history) # 刷新Gradio前端界面 + yield from update_ui_lastest_msg(f'转化PDF编译已经成功, 正在尝试生成对比PDF, 请稍候 ...', chatbot, history) # 刷新Gradio前端界面 result_pdf = pj(work_folder_modified, f'{main_file_modified}.pdf') # get pdf path origin_pdf = pj(work_folder_original, f'{main_file_original}.pdf') # get pdf path if os.path.exists(pj(work_folder, '..', 'translation')): @@ -416,8 +414,11 @@ def 编译Latex(chatbot, history, main_file_original, main_file_modified, work_f from .latex_toolbox import merge_pdfs concat_pdf = pj(work_folder_modified, f'comparison.pdf') merge_pdfs(origin_pdf, result_pdf, concat_pdf) + if os.path.exists(pj(work_folder, '..', 'translation')): + shutil.copyfile(concat_pdf, pj(work_folder, '..', 'translation', 'comparison.pdf')) promote_file_to_downloadzone(concat_pdf, rename_file=None, chatbot=chatbot) # promote file to web UI except Exception as e: + print(e) pass return True # 成功啦 else: diff --git a/crazy_functions/latex_fns/latex_toolbox.py b/crazy_functions/latex_fns/latex_toolbox.py index 4555ff18..0a6a873b 100644 --- a/crazy_functions/latex_fns/latex_toolbox.py +++ b/crazy_functions/latex_fns/latex_toolbox.py @@ -493,11 +493,38 @@ def compile_latex_with_timeout(command, cwd, timeout=60): return False return True +def run_in_subprocess_wrapper_func(func, args, kwargs, return_dict, exception_dict): + import sys + try: + result = func(*args, **kwargs) + return_dict['result'] = result + except Exception as e: + exc_info = sys.exc_info() + exception_dict['exception'] = exc_info +def run_in_subprocess(func): + import multiprocessing + def wrapper(*args, **kwargs): + return_dict = multiprocessing.Manager().dict() + exception_dict = multiprocessing.Manager().dict() + process = multiprocessing.Process(target=run_in_subprocess_wrapper_func, + args=(func, args, kwargs, return_dict, exception_dict)) + process.start() + process.join() + process.close() + if 'exception' in exception_dict: + # ooops, the subprocess ran into an exception + exc_info = exception_dict['exception'] + raise exc_info[1].with_traceback(exc_info[2]) + if 'result' in return_dict.keys(): + # If the subprocess ran successfully, return the result + return return_dict['result'] + return wrapper -def merge_pdfs(pdf1_path, pdf2_path, output_path): - import PyPDF2 +def _merge_pdfs(pdf1_path, pdf2_path, output_path): + import PyPDF2 # PyPDF2这个库有严重的内存泄露问题，把它放到子进程中运行，从而方便内存的释放 Percent = 0.95 + # raise RuntimeError('PyPDF2 has a serious memory leak problem, please use other tools to merge PDF files.') # Open the first PDF file with open(pdf1_path, 'rb') as pdf1_file: pdf1_reader = PyPDF2.PdfFileReader(pdf1_file) @@ -531,3 +558,5 @@ def merge_pdfs(pdf1_path, pdf2_path, output_path): # Save the merged PDF file with open(output_path, 'wb') as output_file: output_writer.write(output_file) + +merge_pdfs = run_in_subprocess(_merge_pdfs) # PyPDF2这个库有严重的内存泄露问题，把它放到子进程中运行，从而方便内存的释放 diff --git a/crazy_functions/multi_stage/multi_stage_utils.py b/crazy_functions/multi_stage/multi_stage_utils.py index 60f07783..1395e79f 100644 --- a/crazy_functions/multi_stage/multi_stage_utils.py +++ b/crazy_functions/multi_stage/multi_stage_utils.py @@ -1,6 +1,7 @@ from pydantic import BaseModel, Field from typing import List from toolbox import update_ui_lastest_msg, disable_auto_promotion +from toolbox import CatchException, update_ui, get_conf, select_api_key, get_log_folder from request_llms.bridge_all import predict_no_ui_long_connection from crazy_functions.json_fns.pydantic_io import GptJsonIO, JsonStringError import time @@ -21,11 +22,7 @@ class GptAcademicState(): def reset(self): pass - def lock_plugin(self, chatbot): - chatbot._cookies['plugin_state'] = pickle.dumps(self) - - def unlock_plugin(self, chatbot): - self.reset() + def dump_state(self, chatbot): chatbot._cookies['plugin_state'] = pickle.dumps(self) def set_state(self, chatbot, key, value): @@ -40,6 +37,57 @@ class GptAcademicState(): state.chatbot = chatbot return state -class GatherMaterials(): - def __init__(self, materials) -> None: - materials = ['image', 'prompt'] \ No newline at end of file + +class GptAcademicGameBaseState(): + """ + 1. first init: __init__ -> + """ + def init_game(self, chatbot, lock_plugin): + self.plugin_name = None + self.callback_fn = None + self.delete_game = False + self.step_cnt = 0 + + def lock_plugin(self, chatbot): + if self.callback_fn is None: + raise ValueError("callback_fn is None") + chatbot._cookies['lock_plugin'] = self.callback_fn + self.dump_state(chatbot) + + def get_plugin_name(self): + if self.plugin_name is None: + raise ValueError("plugin_name is None") + return self.plugin_name + + def dump_state(self, chatbot): + chatbot._cookies[f'plugin_state/{self.get_plugin_name()}'] = pickle.dumps(self) + + def set_state(self, chatbot, key, value): + setattr(self, key, value) + chatbot._cookies[f'plugin_state/{self.get_plugin_name()}'] = pickle.dumps(self) + + @staticmethod + def sync_state(chatbot, llm_kwargs, cls, plugin_name, callback_fn, lock_plugin=True): + state = chatbot._cookies.get(f'plugin_state/{plugin_name}', None) + if state is not None: + state = pickle.loads(state) + else: + state = cls() + state.init_game(chatbot, lock_plugin) + state.plugin_name = plugin_name + state.llm_kwargs = llm_kwargs + state.chatbot = chatbot + state.callback_fn = callback_fn + return state + + def continue_game(self, prompt, chatbot, history): + # 游戏主体 + yield from self.step(prompt, chatbot, history) + self.step_cnt += 1 + # 保存状态，收尾 + self.dump_state(chatbot) + # 如果游戏结束，清理 + if self.delete_game: + chatbot._cookies['lock_plugin'] = None + chatbot._cookies[f'plugin_state/{self.get_plugin_name()}'] = None + yield from update_ui(chatbot=chatbot, history=history) diff --git a/crazy_functions/pdf_fns/breakdown_txt.py b/crazy_functions/pdf_fns/breakdown_txt.py new file mode 100644 index 00000000..a9614814 --- /dev/null +++ b/crazy_functions/pdf_fns/breakdown_txt.py @@ -0,0 +1,125 @@ +from crazy_functions.ipc_fns.mp import run_in_subprocess_with_timeout + +def force_breakdown(txt, limit, get_token_fn): + """ 当无法用标点、空行分割时，我们用最暴力的方法切割 + """ + for i in reversed(range(len(txt))): + if get_token_fn(txt[:i]) < limit: + return txt[:i], txt[i:] + return "Tiktoken未知错误", "Tiktoken未知错误" + + +def maintain_storage(remain_txt_to_cut, remain_txt_to_cut_storage): + """ 为了加速计算，我们采样一个特殊的手段。当 remain_txt_to_cut > `_max` 时， 我们把 _max 后的文字转存至 remain_txt_to_cut_storage + 当 remain_txt_to_cut < `_min` 时，我们再把 remain_txt_to_cut_storage 中的部分文字取出 + """ + _min = int(5e4) + _max = int(1e5) + # print(len(remain_txt_to_cut), len(remain_txt_to_cut_storage)) + if len(remain_txt_to_cut) < _min and len(remain_txt_to_cut_storage) > 0: + remain_txt_to_cut = remain_txt_to_cut + remain_txt_to_cut_storage + remain_txt_to_cut_storage = "" + if len(remain_txt_to_cut) > _max: + remain_txt_to_cut_storage = remain_txt_to_cut[_max:] + remain_txt_to_cut_storage + remain_txt_to_cut = remain_txt_to_cut[:_max] + return remain_txt_to_cut, remain_txt_to_cut_storage + + +def cut(limit, get_token_fn, txt_tocut, must_break_at_empty_line, break_anyway=False): + """ 文本切分 + """ + res = [] + total_len = len(txt_tocut) + fin_len = 0 + remain_txt_to_cut = txt_tocut + remain_txt_to_cut_storage = "" + # 为了加速计算，我们采样一个特殊的手段。当 remain_txt_to_cut > `_max` 时， 我们把 _max 后的文字转存至 remain_txt_to_cut_storage + remain_txt_to_cut, remain_txt_to_cut_storage = maintain_storage(remain_txt_to_cut, remain_txt_to_cut_storage) + + while True: + if get_token_fn(remain_txt_to_cut) <= limit: + # 如果剩余文本的token数小于限制，那么就不用切了 + res.append(remain_txt_to_cut); fin_len+=len(remain_txt_to_cut) + break + else: + # 如果剩余文本的token数大于限制，那么就切 + lines = remain_txt_to_cut.split('\n') + + # 估计一个切分点 + estimated_line_cut = limit / get_token_fn(remain_txt_to_cut) * len(lines) + estimated_line_cut = int(estimated_line_cut) + + # 开始查找合适切分点的偏移（cnt） + cnt = 0 + for cnt in reversed(range(estimated_line_cut)): + if must_break_at_empty_line: + # 首先尝试用双空行（\n\n）作为切分点 + if lines[cnt] != "": + continue + prev = "\n".join(lines[:cnt]) + post = "\n".join(lines[cnt:]) + if get_token_fn(prev) < limit: + break + + if cnt == 0: + # 如果没有找到合适的切分点 + if break_anyway: + # 是否允许暴力切分 + prev, post = force_breakdown(txt_tocut, limit, get_token_fn) + else: + # 不允许直接报错 + raise RuntimeError(f"存在一行极长的文本！{txt_tocut}") + + # 追加列表 + res.append(prev); fin_len+=len(prev) + # 准备下一次迭代 + remain_txt_to_cut = post + remain_txt_to_cut, remain_txt_to_cut_storage = maintain_storage(remain_txt_to_cut, remain_txt_to_cut_storage) + process = fin_len/total_len + print(f'正在文本切分 {int(process*100)}%') + if len(remain_txt_to_cut.strip()) == 0: + break + return res + + +def breakdown_text_to_satisfy_token_limit_(txt, limit, llm_model="gpt-3.5-turbo"): + """ 使用多种方式尝试切分文本，以满足 token 限制 + """ + from request_llms.bridge_all import model_info + enc = model_info[llm_model]['tokenizer'] + def get_token_fn(txt): return len(enc.encode(txt, disallowed_special=())) + try: + # 第1次尝试，将双空行（\n\n）作为切分点 + return cut(limit, get_token_fn, txt, must_break_at_empty_line=True) + except RuntimeError: + try: + # 第2次尝试，将单空行（\n）作为切分点 + return cut(limit, get_token_fn, txt, must_break_at_empty_line=False) + except RuntimeError: + try: + # 第3次尝试，将英文句号（.）作为切分点 + res = cut(limit, get_token_fn, txt.replace('.', '。\n'), must_break_at_empty_line=False) # 这个中文的句号是故意的，作为一个标识而存在 + return [r.replace('。\n', '.') for r in res] + except RuntimeError as e: + try: + # 第4次尝试，将中文句号（。）作为切分点 + res = cut(limit, get_token_fn, txt.replace('。', '。。\n'), must_break_at_empty_line=False) + return [r.replace('。。\n', '。') for r in res] + except RuntimeError as e: + # 第5次尝试，没办法了，随便切一下吧 + return cut(limit, get_token_fn, txt, must_break_at_empty_line=False, break_anyway=True) + +breakdown_text_to_satisfy_token_limit = run_in_subprocess_with_timeout(breakdown_text_to_satisfy_token_limit_, timeout=60) + +if __name__ == '__main__': + from crazy_functions.crazy_utils import read_and_clean_pdf_text + file_content, page_one = read_and_clean_pdf_text("build/assets/at.pdf") + + from request_llms.bridge_all import model_info + for i in range(5): + file_content += file_content + + print(len(file_content)) + TOKEN_LIMIT_PER_FRAGMENT = 2500 + res = breakdown_text_to_satisfy_token_limit(file_content, TOKEN_LIMIT_PER_FRAGMENT) + diff --git a/crazy_functions/pdf_fns/parse_pdf.py b/crazy_functions/pdf_fns/parse_pdf.py index 51f8811f..fa27de51 100644 --- a/crazy_functions/pdf_fns/parse_pdf.py +++ b/crazy_functions/pdf_fns/parse_pdf.py @@ -74,7 +74,7 @@ def produce_report_markdown(gpt_response_collection, meta, paper_meta_info, chat def translate_pdf(article_dict, llm_kwargs, chatbot, fp, generated_conclusion_files, TOKEN_LIMIT_PER_FRAGMENT, DST_LANG): from crazy_functions.pdf_fns.report_gen_html import construct_html - from crazy_functions.crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf + from crazy_functions.pdf_fns.breakdown_txt import breakdown_text_to_satisfy_token_limit from crazy_functions.crazy_utils import request_gpt_model_in_new_thread_with_ui_alive from crazy_functions.crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency @@ -116,7 +116,7 @@ def translate_pdf(article_dict, llm_kwargs, chatbot, fp, generated_conclusion_fi # find a smooth token limit to achieve even seperation count = int(math.ceil(raw_token_num / TOKEN_LIMIT_PER_FRAGMENT)) token_limit_smooth = raw_token_num // count + count - return breakdown_txt_to_satisfy_token_limit_for_pdf(txt, get_token_fn=get_token_num, limit=token_limit_smooth) + return breakdown_text_to_satisfy_token_limit(txt, limit=token_limit_smooth, llm_model=llm_kwargs['llm_model']) for section in article_dict.get('sections'): if len(section['text']) == 0: continue diff --git a/crazy_functions/vector_fns/__init__.py b/crazy_functions/vector_fns/__init__.py new file mode 100644 index 00000000..e69de29b diff --git a/crazy_functions/vector_fns/general_file_loader.py b/crazy_functions/vector_fns/general_file_loader.py new file mode 100644 index 00000000..a512c483 --- /dev/null +++ b/crazy_functions/vector_fns/general_file_loader.py @@ -0,0 +1,70 @@ +# From project chatglm-langchain + + +from langchain.document_loaders import UnstructuredFileLoader +from langchain.text_splitter import CharacterTextSplitter +import re +from typing import List + +class ChineseTextSplitter(CharacterTextSplitter): + def __init__(self, pdf: bool = False, sentence_size: int = None, **kwargs): + super().__init__(**kwargs) + self.pdf = pdf + self.sentence_size = sentence_size + + def split_text1(self, text: str) -> List[str]: + if self.pdf: + text = re.sub(r"\n{3,}", "\n", text) + text = re.sub('\s', ' ', text) + text = text.replace("\n\n", "") + sent_sep_pattern = re.compile('([﹒﹔﹖﹗．。！？]["’”」』]{0,2}|(?=["‘“「『]{1,2}|$))') # del ：； + sent_list = [] + for ele in sent_sep_pattern.split(text): + if sent_sep_pattern.match(ele) and sent_list: + sent_list[-1] += ele + elif ele: + sent_list.append(ele) + return sent_list + + def split_text(self, text: str) -> List[str]: ##此处需要进一步优化逻辑 + if self.pdf: + text = re.sub(r"\n{3,}", r"\n", text) + text = re.sub('\s', " ", text) + text = re.sub("\n\n", "", text) + + text = re.sub(r'([;；.!?。！？\?])([^”’])', r"\1\n\2", text) # 单字符断句符 + text = re.sub(r'(\.{6})([^"’”」』])', r"\1\n\2", text) # 英文省略号 + text = re.sub(r'(\…{2})([^"’”」』])', r"\1\n\2", text) # 中文省略号 + text = re.sub(r'([;；!?。！？\?]["’”」』]{0,2})([^;；!?，。！？\?])', r'\1\n\2', text) + # 如果双引号前有终止符，那么双引号才是句子的终点，把分句符\n放到双引号后，注意前面的几句都小心保留了双引号 + text = text.rstrip() # 段尾如果有多余的\n就去掉它 + # 很多规则中会考虑分号;，但是这里我把它忽略不计，破折号、英文双引号等同样忽略，需要的再做些简单调整即可。 + ls = [i for i in text.split("\n") if i] + for ele in ls: + if len(ele) > self.sentence_size: + ele1 = re.sub(r'([,，.]["’”」』]{0,2})([^,，.])', r'\1\n\2', ele) + ele1_ls = ele1.split("\n") + for ele_ele1 in ele1_ls: + if len(ele_ele1) > self.sentence_size: + ele_ele2 = re.sub(r'([\n]{1,}| {2,}["’”」』]{0,2})([^\s])', r'\1\n\2', ele_ele1) + ele2_ls = ele_ele2.split("\n") + for ele_ele2 in ele2_ls: + if len(ele_ele2) > self.sentence_size: + ele_ele3 = re.sub('( ["’”」』]{0,2})([^ ])', r'\1\n\2', ele_ele2) + ele2_id = ele2_ls.index(ele_ele2) + ele2_ls = ele2_ls[:ele2_id] + [i for i in ele_ele3.split("\n") if i] + ele2_ls[ + ele2_id + 1:] + ele_id = ele1_ls.index(ele_ele1) + ele1_ls = ele1_ls[:ele_id] + [i for i in ele2_ls if i] + ele1_ls[ele_id + 1:] + + id = ls.index(ele) + ls = ls[:id] + [i for i in ele1_ls if i] + ls[id + 1:] + return ls + +def load_file(filepath, sentence_size): + loader = UnstructuredFileLoader(filepath, mode="elements") + textsplitter = ChineseTextSplitter(pdf=False, sentence_size=sentence_size) + docs = loader.load_and_split(text_splitter=textsplitter) + # write_check_file(filepath, docs) + return docs + diff --git a/crazy_functions/vector_fns/vector_database.py b/crazy_functions/vector_fns/vector_database.py new file mode 100644 index 00000000..cffa22cf --- /dev/null +++ b/crazy_functions/vector_fns/vector_database.py @@ -0,0 +1,338 @@ +# From project chatglm-langchain + +import threading +from toolbox import Singleton +import os +import shutil +import os +import uuid +import tqdm +from langchain.vectorstores import FAISS +from langchain.docstore.document import Document +from typing import List, Tuple +import numpy as np +from crazy_functions.vector_fns.general_file_loader import load_file + +embedding_model_dict = { + "ernie-tiny": "nghuyong/ernie-3.0-nano-zh", + "ernie-base": "nghuyong/ernie-3.0-base-zh", + "text2vec-base": "shibing624/text2vec-base-chinese", + "text2vec": "GanymedeNil/text2vec-large-chinese", +} + +# Embedding model name +EMBEDDING_MODEL = "text2vec" + +# Embedding running device +EMBEDDING_DEVICE = "cpu" + +# 基于上下文的prompt模版，请务必保留"{question}"和"{context}" +PROMPT_TEMPLATE = """已知信息： +{context} + +根据上述已知信息，简洁和专业的来回答用户的问题。如果无法从中得到答案，请说 “根据已知信息无法回答该问题” 或 “没有提供足够的相关信息”，不允许在答案中添加编造成分，答案请使用中文。 问题是：{question}""" + +# 文本分句长度 +SENTENCE_SIZE = 100 + +# 匹配后单段上下文长度 +CHUNK_SIZE = 250 + +# LLM input history length +LLM_HISTORY_LEN = 3 + +# return top-k text chunk from vector store +VECTOR_SEARCH_TOP_K = 5 + +# 知识检索内容相关度 Score, 数值范围约为0-1100，如果为0，则不生效，经测试设置为小于500时，匹配结果更精准 +VECTOR_SEARCH_SCORE_THRESHOLD = 0 + +NLTK_DATA_PATH = os.path.join(os.path.dirname(os.path.dirname(__file__)), "nltk_data") + +FLAG_USER_NAME = uuid.uuid4().hex + +# 是否开启跨域，默认为False，如果需要开启，请设置为True +# is open cross domain +OPEN_CROSS_DOMAIN = False + +def similarity_search_with_score_by_vector( + self, embedding: List[float], k: int = 4 +) -> List[Tuple[Document, float]]: + + def seperate_list(ls: List[int]) -> List[List[int]]: + lists = [] + ls1 = [ls[0]] + for i in range(1, len(ls)): + if ls[i - 1] + 1 == ls[i]: + ls1.append(ls[i]) + else: + lists.append(ls1) + ls1 = [ls[i]] + lists.append(ls1) + return lists + + scores, indices = self.index.search(np.array([embedding], dtype=np.float32), k) + docs = [] + id_set = set() + store_len = len(self.index_to_docstore_id) + for j, i in enumerate(indices[0]): + if i == -1 or 0 < self.score_threshold < scores[0][j]: + # This happens when not enough docs are returned. + continue + _id = self.index_to_docstore_id[i] + doc = self.docstore.search(_id) + if not self.chunk_conent: + if not isinstance(doc, Document): + raise ValueError(f"Could not find document for id {_id}, got {doc}") + doc.metadata["score"] = int(scores[0][j]) + docs.append(doc) + continue + id_set.add(i) + docs_len = len(doc.page_content) + for k in range(1, max(i, store_len - i)): + break_flag = False + for l in [i + k, i - k]: + if 0 <= l < len(self.index_to_docstore_id): + _id0 = self.index_to_docstore_id[l] + doc0 = self.docstore.search(_id0) + if docs_len + len(doc0.page_content) > self.chunk_size: + break_flag = True + break + elif doc0.metadata["source"] == doc.metadata["source"]: + docs_len += len(doc0.page_content) + id_set.add(l) + if break_flag: + break + if not self.chunk_conent: + return docs + if len(id_set) == 0 and self.score_threshold > 0: + return [] + id_list = sorted(list(id_set)) + id_lists = seperate_list(id_list) + for id_seq in id_lists: + for id in id_seq: + if id == id_seq[0]: + _id = self.index_to_docstore_id[id] + doc = self.docstore.search(_id) + else: + _id0 = self.index_to_docstore_id[id] + doc0 = self.docstore.search(_id0) + doc.page_content += " " + doc0.page_content + if not isinstance(doc, Document): + raise ValueError(f"Could not find document for id {_id}, got {doc}") + doc_score = min([scores[0][id] for id in [indices[0].tolist().index(i) for i in id_seq if i in indices[0]]]) + doc.metadata["score"] = int(doc_score) + docs.append(doc) + return docs + + +class LocalDocQA: + llm: object = None + embeddings: object = None + top_k: int = VECTOR_SEARCH_TOP_K + chunk_size: int = CHUNK_SIZE + chunk_conent: bool = True + score_threshold: int = VECTOR_SEARCH_SCORE_THRESHOLD + + def init_cfg(self, + top_k=VECTOR_SEARCH_TOP_K, + ): + + self.llm = None + self.top_k = top_k + + def init_knowledge_vector_store(self, + filepath, + vs_path: str or os.PathLike = None, + sentence_size=SENTENCE_SIZE, + text2vec=None): + loaded_files = [] + failed_files = [] + if isinstance(filepath, str): + if not os.path.exists(filepath): + print("路径不存在") + return None + elif os.path.isfile(filepath): + file = os.path.split(filepath)[-1] + try: + docs = load_file(filepath, SENTENCE_SIZE) + print(f"{file} 已成功加载") + loaded_files.append(filepath) + except Exception as e: + print(e) + print(f"{file} 未能成功加载") + return None + elif os.path.isdir(filepath): + docs = [] + for file in tqdm(os.listdir(filepath), desc="加载文件"): + fullfilepath = os.path.join(filepath, file) + try: + docs += load_file(fullfilepath, SENTENCE_SIZE) + loaded_files.append(fullfilepath) + except Exception as e: + print(e) + failed_files.append(file) + + if len(failed_files) > 0: + print("以下文件未能成功加载：") + for file in failed_files: + print(f"{file}\n") + + else: + docs = [] + for file in filepath: + docs += load_file(file, SENTENCE_SIZE) + print(f"{file} 已成功加载") + loaded_files.append(file) + + if len(docs) > 0: + print("文件加载完毕，正在生成向量库") + if vs_path and os.path.isdir(vs_path): + try: + self.vector_store = FAISS.load_local(vs_path, text2vec) + self.vector_store.add_documents(docs) + except: + self.vector_store = FAISS.from_documents(docs, text2vec) + else: + self.vector_store = FAISS.from_documents(docs, text2vec) # docs 为Document列表 + + self.vector_store.save_local(vs_path) + return vs_path, loaded_files + else: + raise RuntimeError("文件加载失败，请检查文件格式是否正确") + + def get_loaded_file(self, vs_path): + ds = self.vector_store.docstore + return set([ds._dict[k].metadata['source'].split(vs_path)[-1] for k in ds._dict]) + + + # query 查询内容 + # vs_path 知识库路径 + # chunk_conent 是否启用上下文关联 + # score_threshold 搜索匹配score阈值 + # vector_search_top_k 搜索知识库内容条数，默认搜索5条结果 + # chunk_sizes 匹配单段内容的连接上下文长度 + def get_knowledge_based_conent_test(self, query, vs_path, chunk_conent, + score_threshold=VECTOR_SEARCH_SCORE_THRESHOLD, + vector_search_top_k=VECTOR_SEARCH_TOP_K, chunk_size=CHUNK_SIZE, + text2vec=None): + self.vector_store = FAISS.load_local(vs_path, text2vec) + self.vector_store.chunk_conent = chunk_conent + self.vector_store.score_threshold = score_threshold + self.vector_store.chunk_size = chunk_size + + embedding = self.vector_store.embedding_function.embed_query(query) + related_docs_with_score = similarity_search_with_score_by_vector(self.vector_store, embedding, k=vector_search_top_k) + + if not related_docs_with_score: + response = {"query": query, + "source_documents": []} + return response, "" + # prompt = f"{query}. You should answer this question using information from following documents: \n\n" + prompt = f"{query}. 你必须利用以下文档中包含的信息回答这个问题: \n\n---\n\n" + prompt += "\n\n".join([f"({k}): " + doc.page_content for k, doc in enumerate(related_docs_with_score)]) + prompt += "\n\n---\n\n" + prompt = prompt.encode('utf-8', 'ignore').decode() # avoid reading non-utf8 chars + # print(prompt) + response = {"query": query, "source_documents": related_docs_with_score} + return response, prompt + + + + +def construct_vector_store(vs_id, vs_path, files, sentence_size, history, one_conent, one_content_segmentation, text2vec): + for file in files: + assert os.path.exists(file), "输入文件不存在：" + file + import nltk + if NLTK_DATA_PATH not in nltk.data.path: nltk.data.path = [NLTK_DATA_PATH] + nltk.data.path + local_doc_qa = LocalDocQA() + local_doc_qa.init_cfg() + filelist = [] + if not os.path.exists(os.path.join(vs_path, vs_id)): + os.makedirs(os.path.join(vs_path, vs_id)) + for file in files: + file_name = file.name if not isinstance(file, str) else file + filename = os.path.split(file_name)[-1] + shutil.copyfile(file_name, os.path.join(vs_path, vs_id, filename)) + filelist.append(os.path.join(vs_path, vs_id, filename)) + vs_path, loaded_files = local_doc_qa.init_knowledge_vector_store(filelist, os.path.join(vs_path, vs_id), sentence_size, text2vec) + + if len(loaded_files): + file_status = f"已添加 {'、'.join([os.path.split(i)[-1] for i in loaded_files if i])} 内容至知识库，并已加载知识库，请开始提问" + else: + pass + # file_status = "文件未成功加载，请重新上传文件" + # print(file_status) + return local_doc_qa, vs_path + +@Singleton +class knowledge_archive_interface(): + def __init__(self) -> None: + self.threadLock = threading.Lock() + self.current_id = "" + self.kai_path = None + self.qa_handle = None + self.text2vec_large_chinese = None + + def get_chinese_text2vec(self): + if self.text2vec_large_chinese is None: + # < -------------------预热文本向量化模组--------------- > + from toolbox import ProxyNetworkActivate + print('Checking Text2vec ...') + from langchain.embeddings.huggingface import HuggingFaceEmbeddings + with ProxyNetworkActivate('Download_LLM'): # 临时地激活代理网络 + self.text2vec_large_chinese = HuggingFaceEmbeddings(model_name="GanymedeNil/text2vec-large-chinese") + + return self.text2vec_large_chinese + + + def feed_archive(self, file_manifest, vs_path, id="default"): + self.threadLock.acquire() + # import uuid + self.current_id = id + self.qa_handle, self.kai_path = construct_vector_store( + vs_id=self.current_id, + vs_path=vs_path, + files=file_manifest, + sentence_size=100, + history=[], + one_conent="", + one_content_segmentation="", + text2vec = self.get_chinese_text2vec(), + ) + self.threadLock.release() + + def get_current_archive_id(self): + return self.current_id + + def get_loaded_file(self, vs_path): + return self.qa_handle.get_loaded_file(vs_path) + + def answer_with_archive_by_id(self, txt, id, vs_path): + self.threadLock.acquire() + if not self.current_id == id: + self.current_id = id + self.qa_handle, self.kai_path = construct_vector_store( + vs_id=self.current_id, + vs_path=vs_path, + files=[], + sentence_size=100, + history=[], + one_conent="", + one_content_segmentation="", + text2vec = self.get_chinese_text2vec(), + ) + VECTOR_SEARCH_SCORE_THRESHOLD = 0 + VECTOR_SEARCH_TOP_K = 4 + CHUNK_SIZE = 512 + resp, prompt = self.qa_handle.get_knowledge_based_conent_test( + query = txt, + vs_path = self.kai_path, + score_threshold=VECTOR_SEARCH_SCORE_THRESHOLD, + vector_search_top_k=VECTOR_SEARCH_TOP_K, + chunk_conent=True, + chunk_size=CHUNK_SIZE, + text2vec = self.get_chinese_text2vec(), + ) + self.threadLock.release() + return resp, prompt \ No newline at end of file diff --git a/crazy_functions/互动小游戏.py b/crazy_functions/互动小游戏.py new file mode 100644 index 00000000..f3786c31 --- /dev/null +++ b/crazy_functions/互动小游戏.py @@ -0,0 +1,40 @@ +from toolbox import CatchException, update_ui, update_ui_lastest_msg +from crazy_functions.multi_stage.multi_stage_utils import GptAcademicGameBaseState +from crazy_functions.crazy_utils import request_gpt_model_in_new_thread_with_ui_alive +from request_llms.bridge_all import predict_no_ui_long_connection +from crazy_functions.game_fns.game_utils import get_code_block, is_same_thing + +@CatchException +def 随机小游戏(prompt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port): + from crazy_functions.game_fns.game_interactive_story import MiniGame_ResumeStory + # 清空历史 + history = [] + # 选择游戏 + cls = MiniGame_ResumeStory + # 如果之前已经初始化了游戏实例，则继续该实例；否则重新初始化 + state = cls.sync_state(chatbot, + llm_kwargs, + cls, + plugin_name='MiniGame_ResumeStory', + callback_fn='crazy_functions.互动小游戏->随机小游戏', + lock_plugin=True + ) + yield from state.continue_game(prompt, chatbot, history) + + +@CatchException +def 随机小游戏1(prompt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port): + from crazy_functions.game_fns.game_ascii_art import MiniGame_ASCII_Art + # 清空历史 + history = [] + # 选择游戏 + cls = MiniGame_ASCII_Art + # 如果之前已经初始化了游戏实例，则继续该实例；否则重新初始化 + state = cls.sync_state(chatbot, + llm_kwargs, + cls, + plugin_name='MiniGame_ASCII_Art', + callback_fn='crazy_functions.互动小游戏->随机小游戏1', + lock_plugin=True + ) + yield from state.continue_game(prompt, chatbot, history) diff --git a/crazy_functions/图片生成.py b/crazy_functions/图片生成.py index 642a9e22..9751e164 100644 --- a/crazy_functions/图片生成.py +++ b/crazy_functions/图片生成.py @@ -2,7 +2,7 @@ from toolbox import CatchException, update_ui, get_conf, select_api_key, get_log from crazy_functions.multi_stage.multi_stage_utils import GptAcademicState -def gen_image(llm_kwargs, prompt, resolution="1024x1024", model="dall-e-2", quality=None): +def gen_image(llm_kwargs, prompt, resolution="1024x1024", model="dall-e-2", quality=None, style=None): import requests, json, time, os from request_llms.bridge_all import model_info @@ -25,7 +25,10 @@ def gen_image(llm_kwargs, prompt, resolution="1024x1024", model="dall-e-2", qual 'model': model, 'response_format': 'url' } - if quality is not None: data.update({'quality': quality}) + if quality is not None: + data['quality'] = quality + if style is not None: + data['style'] = style response = requests.post(url, headers=headers, json=data, proxies=proxies) print(response.content) try: @@ -54,19 +57,25 @@ def edit_image(llm_kwargs, prompt, image_path, resolution="1024x1024", model="da img_endpoint = chat_endpoint.replace('chat/completions','images/edits') # # Generate the image url = img_endpoint + n = 1 headers = { 'Authorization': f"Bearer {api_key}", - 'Content-Type': 'application/json' } - data = { - 'image': open(image_path, 'rb'), - 'prompt': prompt, - 'n': 1, - 'size': resolution, - 'model': model, - 'response_format': 'url' - } - response = requests.post(url, headers=headers, json=data, proxies=proxies) + make_transparent(image_path, image_path+'.tsp.png') + make_square_image(image_path+'.tsp.png', image_path+'.tspsq.png') + resize_image(image_path+'.tspsq.png', image_path+'.ready.png', max_size=1024) + image_path = image_path+'.ready.png' + with open(image_path, 'rb') as f: + file_content = f.read() + files = { + 'image': (os.path.basename(image_path), file_content), + # 'mask': ('mask.png', open('mask.png', 'rb')) + 'prompt': (None, prompt), + "n": (None, str(n)), + 'size': (None, resolution), + } + + response = requests.post(url, headers=headers, files=files, proxies=proxies) print(response.content) try: image_url = json.loads(response.content.decode('utf8'))['data'][0]['url'] @@ -95,7 +104,11 @@ def 图片生成_DALLE2(prompt, llm_kwargs, plugin_kwargs, chatbot, history, sys 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(("您正在调用“图像生成”插件。", "[Local Message] 生成图像, 请先把模型切换至gpt-*或者api2d-*。如果中文Prompt效果不理想, 请尝试英文Prompt。正在处理中 .....")) + if prompt.strip() == "": + chatbot.append((prompt, "[Local Message] 图像生成提示为空白，请在“输入区”输入图像生成提示。")) + yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 界面更新 + return + chatbot.append(("您正在调用“图像生成”插件。", "[Local Message] 生成图像, 请先把模型切换至gpt-*。如果中文Prompt效果不理想, 请尝试英文Prompt。正在处理中 .....")) 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 if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg") resolution = plugin_kwargs.get("advanced_arg", '1024x1024') @@ -112,16 +125,25 @@ def 图片生成_DALLE2(prompt, llm_kwargs, plugin_kwargs, chatbot, history, sys @CatchException def 图片生成_DALLE3(prompt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port): history = [] # 清空历史,以免输入溢出 - chatbot.append(("您正在调用“图像生成”插件。", "[Local Message] 生成图像, 请先把模型切换至gpt-*或者api2d-*。如果中文Prompt效果不理想, 请尝试英文Prompt。正在处理中 .....")) + if prompt.strip() == "": + chatbot.append((prompt, "[Local Message] 图像生成提示为空白，请在“输入区”输入图像生成提示。")) + yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 界面更新 + return + chatbot.append(("您正在调用“图像生成”插件。", "[Local Message] 生成图像, 请先把模型切换至gpt-*。如果中文Prompt效果不理想, 请尝试英文Prompt。正在处理中 .....")) 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 if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg") - resolution = plugin_kwargs.get("advanced_arg", '1024x1024').lower() - if resolution.endswith('-hd'): - resolution = resolution.replace('-hd', '') - quality = 'hd' - else: - quality = 'standard' - image_url, image_path = gen_image(llm_kwargs, prompt, resolution, model="dall-e-3", quality=quality) + resolution_arg = plugin_kwargs.get("advanced_arg", '1024x1024-standard-vivid').lower() + parts = resolution_arg.split('-') + resolution = parts[0] # 解析分辨率 + quality = 'standard' # 质量与风格默认值 + style = 'vivid' + # 遍历检查是否有额外参数 + for part in parts[1:]: + if part in ['hd', 'standard']: + quality = part + elif part in ['vivid', 'natural']: + style = part + image_url, image_path = gen_image(llm_kwargs, prompt, resolution, model="dall-e-3", quality=quality, style=style) chatbot.append([prompt, f'图像中转网址:
`{image_url}`
'+ f'中转网址预览:
' @@ -130,6 +152,7 @@ def 图片生成_DALLE3(prompt, llm_kwargs, plugin_kwargs, chatbot, history, sys ]) yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 界面更新 + class ImageEditState(GptAcademicState): # 尚未完成 def get_image_file(self, x): @@ -142,18 +165,27 @@ class ImageEditState(GptAcademicState): file = None if not confirm else file_manifest[0] return confirm, file + def lock_plugin(self, chatbot): + chatbot._cookies['lock_plugin'] = 'crazy_functions.图片生成->图片修改_DALLE2' + self.dump_state(chatbot) + + def unlock_plugin(self, chatbot): + self.reset() + chatbot._cookies['lock_plugin'] = None + self.dump_state(chatbot) + def get_resolution(self, x): return (x in ['256x256', '512x512', '1024x1024']), x - + def get_prompt(self, x): confirm = (len(x)>=5) and (not self.get_resolution(x)[0]) and (not self.get_image_file(x)[0]) return confirm, x - + def reset(self): self.req = [ - {'value':None, 'description': '请先上传图像（必须是.png格式）, 然后再次点击本插件', 'verify_fn': self.get_image_file}, - {'value':None, 'description': '请输入分辨率,可选：256x256, 512x512 或 1024x1024', 'verify_fn': self.get_resolution}, - {'value':None, 'description': '请输入修改需求,建议您使用英文提示词', 'verify_fn': self.get_prompt}, + {'value':None, 'description': '请先上传图像（必须是.png格式）, 然后再次点击本插件', 'verify_fn': self.get_image_file}, + {'value':None, 'description': '请输入分辨率,可选：256x256, 512x512 或 1024x1024, 然后再次点击本插件', 'verify_fn': self.get_resolution}, + {'value':None, 'description': '请输入修改需求,建议您使用英文提示词, 然后再次点击本插件', 'verify_fn': self.get_prompt}, ] self.info = "" @@ -163,7 +195,7 @@ class ImageEditState(GptAcademicState): confirm, res = r['verify_fn'](prompt) if confirm: r['value'] = res - self.set_state(chatbot, 'dummy_key', 'dummy_value') + self.dump_state(chatbot) break return self @@ -182,23 +214,63 @@ def 图片修改_DALLE2(prompt, llm_kwargs, plugin_kwargs, chatbot, history, sys history = [] # 清空历史 state = ImageEditState.get_state(chatbot, ImageEditState) state = state.feed(prompt, chatbot) + state.lock_plugin(chatbot) if not state.already_obtained_all_materials(): - chatbot.append(["图片修改（先上传图片,再输入修改需求,最后输入分辨率）", state.next_req()]) + chatbot.append(["图片修改\n\n1. 上传图片（图片中需要修改的位置用橡皮擦擦除为纯白色，即RGB=255,255,255）\n2. 输入分辨率 \n3. 输入修改需求", state.next_req()]) yield from update_ui(chatbot=chatbot, history=history) return - image_path = state.req[0] - resolution = state.req[1] - prompt = state.req[2] + image_path = state.req[0]['value'] + resolution = state.req[1]['value'] + prompt = state.req[2]['value'] chatbot.append(["图片修改, 执行中", f"图片:`{image_path}`
分辨率:`{resolution}`
修改需求:`{prompt}`"]) yield from update_ui(chatbot=chatbot, history=history) - image_url, image_path = edit_image(llm_kwargs, prompt, image_path, resolution) - chatbot.append([state.prompt, + chatbot.append([prompt, f'图像中转网址:
`{image_url}`
'+ f'中转网址预览:
' f'本地文件地址:
`{image_path}`
'+ f'本地文件预览:
' ]) yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 界面更新 + state.unlock_plugin(chatbot) +def make_transparent(input_image_path, output_image_path): + from PIL import Image + image = Image.open(input_image_path) + image = image.convert("RGBA") + data = image.getdata() + new_data = [] + for item in data: + if item[0] == 255 and item[1] == 255 and item[2] == 255: + new_data.append((255, 255, 255, 0)) + else: + new_data.append(item) + image.putdata(new_data) + image.save(output_image_path, "PNG") + +def resize_image(input_path, output_path, max_size=1024): + from PIL import Image + with Image.open(input_path) as img: + width, height = img.size + if width > max_size or height > max_size: + if width >= height: + new_width = max_size + new_height = int((max_size / width) * height) + else: + new_height = max_size + new_width = int((max_size / height) * width) + + resized_img = img.resize(size=(new_width, new_height)) + resized_img.save(output_path) + else: + img.save(output_path) + +def make_square_image(input_path, output_path): + from PIL import Image + with Image.open(input_path) as img: + width, height = img.size + size = max(width, height) + new_img = Image.new("RGBA", (size, size), color="black") + new_img.paste(img, ((size - width) // 2, (size - height) // 2)) + new_img.save(output_path) diff --git a/crazy_functions/总结word文档.py b/crazy_functions/总结word文档.py index b3923071..01ee1e6b 100644 --- a/crazy_functions/总结word文档.py +++ b/crazy_functions/总结word文档.py @@ -29,17 +29,12 @@ def 解析docx(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot except: raise RuntimeError('请先将.doc文档转换为.docx文档。') - print(file_content) # 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 - from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf + from crazy_functions.pdf_fns.breakdown_txt import breakdown_text_to_satisfy_token_limit from request_llms.bridge_all import model_info max_token = model_info[llm_kwargs['llm_model']]['max_token'] TOKEN_LIMIT_PER_FRAGMENT = max_token * 3 // 4 - paper_fragments = breakdown_txt_to_satisfy_token_limit_for_pdf( - txt=file_content, - get_token_fn=model_info[llm_kwargs['llm_model']]['token_cnt'], - limit=TOKEN_LIMIT_PER_FRAGMENT - ) + paper_fragments = breakdown_text_to_satisfy_token_limit(txt=file_content, limit=TOKEN_LIMIT_PER_FRAGMENT, llm_model=llm_kwargs['llm_model']) this_paper_history = [] for i, paper_frag in enumerate(paper_fragments): i_say = f'请对下面的文章片段用中文做概述，文件名是{os.path.relpath(fp, project_folder)}，文章内容是 ```{paper_frag}```' diff --git a/crazy_functions/批量Markdown翻译.py b/crazy_functions/批量Markdown翻译.py index 12b4ef09..8665d6df 100644 --- a/crazy_functions/批量Markdown翻译.py +++ b/crazy_functions/批量Markdown翻译.py @@ -28,8 +28,8 @@ class PaperFileGroup(): 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) + from crazy_functions.pdf_fns.breakdown_txt import breakdown_text_to_satisfy_token_limit + segments = breakdown_text_to_satisfy_token_limit(file_content, max_token_limit) for j, segment in enumerate(segments): self.sp_file_contents.append(segment) self.sp_file_index.append(index) diff --git a/crazy_functions/批量总结PDF文档.py b/crazy_functions/批量总结PDF文档.py index 7fc3e415..e289c47b 100644 --- a/crazy_functions/批量总结PDF文档.py +++ b/crazy_functions/批量总结PDF文档.py @@ -20,14 +20,9 @@ def 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, TOKEN_LIMIT_PER_FRAGMENT = 2500 - from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf - from request_llms.bridge_all import model_info - enc = model_info["gpt-3.5-turbo"]['tokenizer'] - def get_token_num(txt): return len(enc.encode(txt, disallowed_special=())) - 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) + from crazy_functions.pdf_fns.breakdown_txt import breakdown_text_to_satisfy_token_limit + paper_fragments = breakdown_text_to_satisfy_token_limit(txt=file_content, limit=TOKEN_LIMIT_PER_FRAGMENT, llm_model=llm_kwargs['llm_model']) + page_one_fragments = breakdown_text_to_satisfy_token_limit(txt=str(page_one), limit=TOKEN_LIMIT_PER_FRAGMENT//4, llm_model=llm_kwargs['llm_model']) # 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] diff --git a/crazy_functions/批量翻译PDF文档_多线程.py b/crazy_functions/批量翻译PDF文档_多线程.py index 73cf5920..a1f0f312 100644 --- a/crazy_functions/批量翻译PDF文档_多线程.py +++ b/crazy_functions/批量翻译PDF文档_多线程.py @@ -91,14 +91,9 @@ def 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, page_one = str(page_one).encode('utf-8', 'ignore').decode() # avoid reading non-utf8 chars # 递归地切割PDF文件 - from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf - from request_llms.bridge_all import model_info - enc = model_info["gpt-3.5-turbo"]['tokenizer'] - def get_token_num(txt): return len(enc.encode(txt, disallowed_special=())) - 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=page_one, get_token_fn=get_token_num, limit=TOKEN_LIMIT_PER_FRAGMENT//4) + from crazy_functions.pdf_fns.breakdown_txt import breakdown_text_to_satisfy_token_limit + paper_fragments = breakdown_text_to_satisfy_token_limit(txt=file_content, limit=TOKEN_LIMIT_PER_FRAGMENT, llm_model=llm_kwargs['llm_model']) + page_one_fragments = breakdown_text_to_satisfy_token_limit(txt=page_one, limit=TOKEN_LIMIT_PER_FRAGMENT//4, llm_model=llm_kwargs['llm_model']) # 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] diff --git a/crazy_functions/理解PDF文档内容.py b/crazy_functions/理解PDF文档内容.py index ef967889..439d78ea 100644 --- a/crazy_functions/理解PDF文档内容.py +++ b/crazy_functions/理解PDF文档内容.py @@ -18,14 +18,9 @@ def 解析PDF(file_name, llm_kwargs, plugin_kwargs, chatbot, history, system_pro TOKEN_LIMIT_PER_FRAGMENT = 2500 - from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf - from request_llms.bridge_all import model_info - enc = model_info["gpt-3.5-turbo"]['tokenizer'] - def get_token_num(txt): return len(enc.encode(txt, disallowed_special=())) - 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) + from crazy_functions.pdf_fns.breakdown_txt import breakdown_text_to_satisfy_token_limit + paper_fragments = breakdown_text_to_satisfy_token_limit(txt=file_content, limit=TOKEN_LIMIT_PER_FRAGMENT, llm_model=llm_kwargs['llm_model']) + page_one_fragments = breakdown_text_to_satisfy_token_limit(txt=str(page_one), limit=TOKEN_LIMIT_PER_FRAGMENT//4, llm_model=llm_kwargs['llm_model']) # 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] @@ -45,7 +40,7 @@ def 解析PDF(file_name, llm_kwargs, plugin_kwargs, chatbot, history, system_pro 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]}" + 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 diff --git a/crazy_functions/Langchain知识库.py b/crazy_functions/知识库问答.py similarity index 69% rename from crazy_functions/Langchain知识库.py rename to crazy_functions/知识库问答.py index 8433895f..e1cd00ca 100644 --- a/crazy_functions/Langchain知识库.py +++ b/crazy_functions/知识库问答.py @@ -1,10 +1,19 @@ -from toolbox import CatchException, update_ui, ProxyNetworkActivate, update_ui_lastest_msg +from toolbox import CatchException, update_ui, ProxyNetworkActivate, update_ui_lastest_msg, get_log_folder, get_user from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive, get_files_from_everything +install_msg =""" +1. python -m pip install torch --index-url https://download.pytorch.org/whl/cpu + +2. python -m pip install transformers protobuf langchain sentence-transformers faiss-cpu nltk beautifulsoup4 bitsandbytes tabulate icetk --upgrade + +3. python -m pip install unstructured[all-docs] --upgrade + +4. python -c 'import nltk; nltk.download("punkt")' +""" @CatchException -def 知识库问答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port): +def 知识库文件注入(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 @@ -25,15 +34,15 @@ def 知识库问答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_pro # resolve deps try: - from zh_langchain import construct_vector_store - from langchain.embeddings.huggingface import HuggingFaceEmbeddings - from .crazy_utils import knowledge_archive_interface + # from zh_langchain import construct_vector_store + # from langchain.embeddings.huggingface import HuggingFaceEmbeddings + from crazy_functions.vector_fns.vector_database import knowledge_archive_interface except Exception as e: - chatbot.append(["依赖不足", "导入依赖失败。正在尝试自动安装，请查看终端的输出或耐心等待..."]) + chatbot.append(["依赖不足", f"{str(e)}\n\n导入依赖失败。请用以下命令安装" + install_msg]) yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 - from .crazy_utils import try_install_deps - try_install_deps(['zh_langchain==0.2.1', 'pypinyin'], reload_m=['pypinyin', 'zh_langchain']) - yield from update_ui_lastest_msg("安装完成，您可以再次重试。", chatbot, history) + # from .crazy_utils import try_install_deps + # try_install_deps(['zh_langchain==0.2.1', 'pypinyin'], reload_m=['pypinyin', 'zh_langchain']) + # yield from update_ui_lastest_msg("安装完成，您可以再次重试。", chatbot, history) return # < --------------------读取文件--------------- > @@ -42,7 +51,7 @@ def 知识库问答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_pro for sp in spl: _, file_manifest_tmp, _ = get_files_from_everything(txt, type=f'.{sp}') file_manifest += file_manifest_tmp - + if len(file_manifest) == 0: chatbot.append(["没有找到任何可读取文件", "当前支持的格式包括: txt, md, docx, pptx, pdf, json等"]) yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 @@ -62,13 +71,14 @@ def 知识库问答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_pro print('Establishing knowledge archive ...') with ProxyNetworkActivate('Download_LLM'): # 临时地激活代理网络 kai = knowledge_archive_interface() - kai.feed_archive(file_manifest=file_manifest, id=kai_id) - kai_files = kai.get_loaded_file() + vs_path = get_log_folder(user=get_user(chatbot), plugin_name='vec_store') + kai.feed_archive(file_manifest=file_manifest, vs_path=vs_path, id=kai_id) + kai_files = kai.get_loaded_file(vs_path=vs_path) kai_files = '
'.join(kai_files) # chatbot.append(['知识库构建成功', "正在将知识库存储至cookie中"]) # yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # chatbot._cookies['langchain_plugin_embedding'] = kai.get_current_archive_id() - # chatbot._cookies['lock_plugin'] = 'crazy_functions.Langchain知识库->读取知识库作答' + # chatbot._cookies['lock_plugin'] = 'crazy_functions.知识库文件注入->读取知识库作答' # chatbot.append(['完成', "“根据知识库作答”函数插件已经接管问答系统, 提问吧! 但注意, 您接下来不能再使用其他插件了，刷新页面即可以退出知识库问答模式。"]) chatbot.append(['构建完成', f"当前知识库内的有效文件：\n\n---\n\n{kai_files}\n\n---\n\n请切换至“知识库问答”插件进行知识库访问, 或者使用此插件继续上传更多文件。"]) 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 @@ -77,15 +87,15 @@ def 知识库问答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_pro def 读取知识库作答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port=-1): # resolve deps try: - from zh_langchain import construct_vector_store - from langchain.embeddings.huggingface import HuggingFaceEmbeddings - from .crazy_utils import knowledge_archive_interface + # from zh_langchain import construct_vector_store + # from langchain.embeddings.huggingface import HuggingFaceEmbeddings + from crazy_functions.vector_fns.vector_database import knowledge_archive_interface except Exception as e: - chatbot.append(["依赖不足", "导入依赖失败。正在尝试自动安装，请查看终端的输出或耐心等待..."]) + chatbot.append(["依赖不足", f"{str(e)}\n\n导入依赖失败。请用以下命令安装" + install_msg]) yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 - from .crazy_utils import try_install_deps - try_install_deps(['zh_langchain==0.2.1', 'pypinyin'], reload_m=['pypinyin', 'zh_langchain']) - yield from update_ui_lastest_msg("安装完成，您可以再次重试。", chatbot, history) + # from .crazy_utils import try_install_deps + # try_install_deps(['zh_langchain==0.2.1', 'pypinyin'], reload_m=['pypinyin', 'zh_langchain']) + # yield from update_ui_lastest_msg("安装完成，您可以再次重试。", chatbot, history) return # < ------------------- --------------- > @@ -93,7 +103,8 @@ def 读取知识库作答(txt, llm_kwargs, plugin_kwargs, chatbot, history, syst if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg") kai_id = plugin_kwargs.get("advanced_arg", 'default') - resp, prompt = kai.answer_with_archive_by_id(txt, kai_id) + vs_path = get_log_folder(user=get_user(chatbot), plugin_name='vec_store') + resp, prompt = kai.answer_with_archive_by_id(txt, kai_id, vs_path) chatbot.append((txt, f'[知识库 {kai_id}] ' + prompt)) 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 diff --git a/crazy_functions/解析JupyterNotebook.py b/crazy_functions/解析JupyterNotebook.py index eeccadf7..3c2b5783 100644 --- a/crazy_functions/解析JupyterNotebook.py +++ b/crazy_functions/解析JupyterNotebook.py @@ -12,13 +12,6 @@ class PaperFileGroup(): self.sp_file_index = [] self.sp_file_tag = [] - # count_token - from request_llms.bridge_all import model_info - enc = model_info["gpt-3.5-turbo"]['tokenizer'] - def get_token_num(txt): return len( - enc.encode(txt, disallowed_special=())) - self.get_token_num = get_token_num - def run_file_split(self, max_token_limit=1900): """ 将长文本分离开来 @@ -29,9 +22,8 @@ class PaperFileGroup(): 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) + from crazy_functions.pdf_fns.breakdown_txt import breakdown_text_to_satisfy_token_limit + segments = breakdown_text_to_satisfy_token_limit(file_content, max_token_limit) for j, segment in enumerate(segments): self.sp_file_contents.append(segment) self.sp_file_index.append(index) diff --git a/docs/GithubAction+AllCapacityBeta b/docs/GithubAction+AllCapacityBeta new file mode 100644 index 00000000..d3a06ee1 --- /dev/null +++ b/docs/GithubAction+AllCapacityBeta @@ -0,0 +1,53 @@ +# docker build -t gpt-academic-all-capacity -f docs/GithubAction+AllCapacity --network=host --build-arg http_proxy=http://localhost:10881 --build-arg https_proxy=http://localhost:10881 . +# docker build -t gpt-academic-all-capacity -f docs/GithubAction+AllCapacityBeta --network=host . +# docker run -it --net=host gpt-academic-all-capacity bash + +# 从NVIDIA源，从而支持显卡（检查宿主的nvidia-smi中的cuda版本必须>=11.3） +FROM fuqingxu/11.3.1-runtime-ubuntu20.04-with-texlive:latest + +# use python3 as the system default python +WORKDIR /gpt +RUN curl -sS https://bootstrap.pypa.io/get-pip.py | python3.8 + +# # 非必要步骤，更换pip源 （以下三行，可以删除） +# 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 + +# 下载pytorch +RUN python3 -m pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu113 +# 准备pip依赖 +RUN python3 -m pip install openai numpy arxiv rich +RUN python3 -m pip install colorama Markdown pygments pymupdf +RUN python3 -m pip install python-docx moviepy pdfminer +RUN python3 -m pip install zh_langchain==0.2.1 pypinyin +RUN python3 -m pip install rarfile py7zr +RUN python3 -m pip install aliyun-python-sdk-core==2.13.3 pyOpenSSL webrtcvad scipy git+https://github.com/aliyun/alibabacloud-nls-python-sdk.git +# 下载分支 +WORKDIR /gpt +RUN git clone --depth=1 https://github.com/binary-husky/gpt_academic.git +WORKDIR /gpt/gpt_academic +RUN git clone --depth=1 https://github.com/OpenLMLab/MOSS.git request_llms/moss + +RUN python3 -m pip install -r requirements.txt +RUN python3 -m pip install -r request_llms/requirements_moss.txt +RUN python3 -m pip install -r request_llms/requirements_qwen.txt +RUN python3 -m pip install -r request_llms/requirements_chatglm.txt +RUN python3 -m pip install -r request_llms/requirements_newbing.txt +RUN python3 -m pip install nougat-ocr + +# 预热Tiktoken模块 +RUN python3 -c 'from check_proxy import warm_up_modules; warm_up_modules()' + +# 安装知识库插件的额外依赖 +RUN apt-get update && apt-get install libgl1 -y +RUN pip3 install transformers protobuf langchain sentence-transformers faiss-cpu nltk beautifulsoup4 bitsandbytes tabulate icetk --upgrade +RUN pip3 install unstructured[all-docs] --upgrade +RUN python3 -c 'from check_proxy import warm_up_vectordb; warm_up_vectordb()' +RUN rm -rf /usr/local/lib/python3.8/dist-packages/tests + + +# COPY .cache /root/.cache +# COPY config_private.py config_private.py +# 启动 +CMD ["python3", "-u", "main.py"] diff --git a/docs/GithubAction+NoLocal+Vectordb b/docs/GithubAction+NoLocal+Vectordb new file mode 100644 index 00000000..45074d93 --- /dev/null +++ b/docs/GithubAction+NoLocal+Vectordb @@ -0,0 +1,26 @@ +# 此Dockerfile适用于“无本地模型”的环境构建，如果需要使用chatglm等本地模型，请参考 docs/Dockerfile+ChatGLM +# 如何构建: 先修改 `config.py`， 然后 docker build -t gpt-academic-nolocal-vs -f docs/GithubAction+NoLocal+Vectordb . +# 如何运行: docker run --rm -it --net=host gpt-academic-nolocal-vs +FROM python:3.11 + +# 指定路径 +WORKDIR /gpt + +# 装载项目文件 +COPY . . + +# 安装依赖 +RUN pip3 install -r requirements.txt + +# 安装知识库插件的额外依赖 +RUN apt-get update && apt-get install libgl1 -y +RUN pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cpu +RUN pip3 install transformers protobuf langchain sentence-transformers faiss-cpu nltk beautifulsoup4 bitsandbytes tabulate icetk --upgrade +RUN pip3 install unstructured[all-docs] --upgrade +RUN python3 -c 'from check_proxy import warm_up_vectordb; warm_up_vectordb()' + +# 可选步骤，用于预热模块 +RUN python3 -c 'from check_proxy import warm_up_modules; warm_up_modules()' + +# 启动 +CMD ["python3", "-u", "main.py"] diff --git a/docs/translate_english.json b/docs/translate_english.json index 955dcaf9..3920e1f6 100644 --- a/docs/translate_english.json +++ b/docs/translate_english.json @@ -923,7 +923,7 @@ "的第": "The", "个片段": "fragment", "总结文章": "Summarize the article", - "根据以上的对话": "According to the above dialogue", + "根据以上的对话": "According to the conversation above", "的主要内容": "The main content of", "所有文件都总结完成了吗": "Are all files summarized?", "如果是.doc文件": "If it is a .doc file", @@ -1501,7 +1501,7 @@ "发送请求到OpenAI后": "After sending the request to OpenAI", "上下布局": "Vertical Layout", "左右布局": "Horizontal Layout", - "对话窗的高度": "Height of the Dialogue Window", + "对话窗的高度": "Height of the Conversation Window", "重试的次数限制": "Retry Limit", "gpt4现在只对申请成功的人开放": "GPT-4 is now only open to those who have successfully applied", "提高限制请查询": "Please check for higher limits", @@ -2183,9 +2183,8 @@ "找不到合适插件执行该任务": "Cannot find a suitable plugin to perform this task", "接驳VoidTerminal": "Connect to VoidTerminal", "**很好": "**Very good", - "对话|编程": "Conversation|Programming", - "对话|编程|学术": "Conversation|Programming|Academic", - "4. 建议使用 GPT3.5 或更强的模型": "4. It is recommended to use GPT3.5 or a stronger model", + "对话|编程": "Conversation&ImageGenerating|Programming", + "对话|编程|学术": "Conversation&ImageGenerating|Programming|Academic", "4. 建议使用 GPT3.5 或更强的模型": "4. It is recommended to use GPT3.5 or a stronger model", "「请调用插件翻译PDF论文": "Please call the plugin to translate the PDF paper", "3. 如果您使用「调用插件xxx」、「修改配置xxx」、「请问」等关键词": "3. If you use keywords such as 'call plugin xxx', 'modify configuration xxx', 'please', etc.", "以下是一篇学术论文的基本信息": "The following is the basic information of an academic paper", @@ -2630,7 +2629,7 @@ "已经被记忆": "Already memorized", "默认用英文的": "Default to English", "错误追踪": "Error tracking", - "对话|编程|学术|智能体": "Dialogue|Programming|Academic|Intelligent agent", + "对话&编程|编程|学术|智能体": "Conversation&ImageGenerating|Programming|Academic|Intelligent agent", "请检查": "Please check", "检测到被滞留的缓存文档": "Detected cached documents being left behind", "还有哪些场合允许使用代理": "What other occasions allow the use of proxies", @@ -2864,7 +2863,7 @@ "加载API_KEY": "Loading API_KEY", "协助您编写代码": "Assist you in writing code", "我可以为您提供以下服务": "I can provide you with the following services", - "排队中请稍后 ...": "Please wait in line ...", + "排队中请稍候 ...": "Please wait in line ...", "建议您使用英文提示词": "It is recommended to use English prompts", "不能支撑AutoGen运行": "Cannot support AutoGen operation", "帮助您解决编程问题": "Help you solve programming problems", @@ -2903,5 +2902,107 @@ "高优先级": "High priority", "请配置ZHIPUAI_API_KEY": "Please configure ZHIPUAI_API_KEY", "单个azure模型": "Single Azure model", - "预留参数 context 未实现": "Reserved parameter 'context' not implemented" -} \ No newline at end of file + "预留参数 context 未实现": "Reserved parameter 'context' not implemented", + "在输入区输入临时API_KEY后提交": "Submit after entering temporary API_KEY in the input area", + "鸟": "Bird", + "图片中需要修改的位置用橡皮擦擦除为纯白色": "Erase the areas in the image that need to be modified with an eraser to pure white", + "└── PDF文档精准解析": "└── Accurate parsing of PDF documents", + "└── ALLOW_RESET_CONFIG 是否允许通过自然语言描述修改本页的配置": "└── ALLOW_RESET_CONFIG Whether to allow modifying the configuration of this page through natural language description", + "等待指令": "Waiting for instructions", + "不存在": "Does not exist", + "选择游戏": "Select game", + "本地大模型示意图": "Local large model diagram", + "无视此消息即可": "You can ignore this message", + "即RGB=255": "That is, RGB=255", + "如需追问": "If you have further questions", + "也可以是具体的模型路径": "It can also be a specific model path", + "才会起作用": "Will take effect", + "下载失败": "Download failed", + "网页刷新后失效": "Invalid after webpage refresh", + "crazy_functions.互动小游戏-": "crazy_functions.Interactive mini game-", + "右对齐": "Right alignment", + "您可以调用下拉菜单中的“LoadConversationHistoryArchive”还原当下的对话": "You can use the 'LoadConversationHistoryArchive' in the drop-down menu to restore the current conversation", + "左对齐": "Left alignment", + "使用默认的 FP16": "Use default FP16", + "一小时": "One hour", + "从而方便内存的释放": "Thus facilitating memory release", + "如何临时更换API_KEY": "How to temporarily change API_KEY", + "请输入 1024x1024-HD": "Please enter 1024x1024-HD", + "使用 INT8 量化": "Use INT8 quantization", + "3. 输入修改需求": "3. Enter modification requirements", + "刷新界面 由于请求gpt需要一段时间": "Refreshing the interface takes some time due to the request for gpt", + "随机小游戏": "Random mini game", + "那么请在下面的QWEN_MODEL_SELECTION中指定具体的模型": "So please specify the specific model in QWEN_MODEL_SELECTION below", + "表值": "Table value", + "我画你猜": "I draw, you guess", + "狗": "Dog", + "2. 输入分辨率": "2. Enter resolution", + "鱼": "Fish", + "尚未完成": "Not yet completed", + "表头": "Table header", + "填localhost或者127.0.0.1": "Fill in localhost or 127.0.0.1", + "请上传jpg格式的图片": "Please upload images in jpg format", + "API_URL_REDIRECT填写格式是错误的": "The format of API_URL_REDIRECT is incorrect", + "├── RWKV的支持见Wiki": "Support for RWKV is available in the Wiki", + "如果中文Prompt效果不理想": "If the Chinese prompt is not effective", + "/SEAFILE_LOCAL/50503047/我的资料库/学位/paperlatex/aaai/Fu_8368_with_appendix": "/SEAFILE_LOCAL/50503047/My Library/Degree/paperlatex/aaai/Fu_8368_with_appendix", + "只有当AVAIL_LLM_MODELS包含了对应本地模型时": "Only when AVAIL_LLM_MODELS contains the corresponding local model", + "选择本地模型变体": "Choose the local model variant", + "如果您确信自己没填错": "If you are sure you haven't made a mistake", + "PyPDF2这个库有严重的内存泄露问题": "PyPDF2 library has serious memory leak issues", + "整理文件集合 输出消息": "Organize file collection and output message", + "没有检测到任何近期上传的图像文件": "No recently uploaded image files detected", + "游戏结束": "Game over", + "调用结束": "Call ended", + "猫": "Cat", + "请及时切换模型": "Please switch models in time", + "次中": "In the meantime", + "如需生成高清图像": "If you need to generate high-definition images", + "CPU 模式": "CPU mode", + "项目目录": "Project directory", + "动物": "Animal", + "居中对齐": "Center alignment", + "请注意拓展名需要小写": "Please note that the extension name needs to be lowercase", + "重试第": "Retry", + "实验性功能": "Experimental feature", + "猜错了": "Wrong guess", + "打开你的代理软件查看代理协议": "Open your proxy software to view the proxy agreement", + "您不需要再重复强调该文件的路径了": "You don't need to emphasize the file path again", + "请阅读": "Please read", + "请直接输入您的问题": "Please enter your question directly", + "API_URL_REDIRECT填错了": "API_URL_REDIRECT is filled incorrectly", + "谜底是": "The answer is", + "第一个模型": "The first model", + "你猜对了！": "You guessed it right!", + "已经接收到您上传的文件": "The file you uploaded has been received", + "您正在调用“图像生成”插件": "You are calling the 'Image Generation' plugin", + "刷新界面 界面更新": "Refresh the interface, interface update", + "如果之前已经初始化了游戏实例": "If the game instance has been initialized before", + "文件": "File", + "老鼠": "Mouse", + "列2": "Column 2", + "等待图片": "Waiting for image", + "使用 INT4 量化": "Use INT4 quantization", + "from crazy_functions.互动小游戏 import 随机小游戏": "TranslatedText", + "游戏主体": "TranslatedText", + "该模型不具备上下文对话能力": "TranslatedText", + "列3": "TranslatedText", + "清理": "TranslatedText", + "检查量化配置": "TranslatedText", + "如果游戏结束": "TranslatedText", + "蛇": "TranslatedText", + "则继续该实例；否则重新初始化": "TranslatedText", + "e.g. cat and 猫 are the same thing": "TranslatedText", + "第三个模型": "TranslatedText", + "如果你选择Qwen系列的模型": "TranslatedText", + "列4": "TranslatedText", + "输入“exit”获取答案": "TranslatedText", + "把它放到子进程中运行": "TranslatedText", + "列1": "TranslatedText", + "使用该模型需要额外依赖": "TranslatedText", + "再试试": "TranslatedText", + "1. 上传图片": "TranslatedText", + "保存状态": "TranslatedText", + "GPT-Academic对话存档": "TranslatedText", + "Arxiv论文精细翻译": "TranslatedText" +} diff --git a/docs/translate_traditionalchinese.json b/docs/translate_traditionalchinese.json index 9ca7cbaa..4edc65de 100644 --- a/docs/translate_traditionalchinese.json +++ b/docs/translate_traditionalchinese.json @@ -1043,9 +1043,9 @@ "jittorllms响应异常": "jittorllms response exception", "在项目根目录运行这两个指令": "Run these two commands in the project root directory", "获取tokenizer": "Get tokenizer", - "chatbot 为WebUI中显示的对话列表": "chatbot is the list of dialogues displayed in WebUI", + "chatbot 为WebUI中显示的对话列表": "chatbot is the list of conversations displayed in WebUI", "test_解析一个Cpp项目": "test_parse a Cpp project", - "将对话记录history以Markdown格式写入文件中": "Write the dialogue record history to a file in Markdown format", + "将对话记录history以Markdown格式写入文件中": "Write the conversations record history to a file in Markdown format", "装饰器函数": "Decorator function", "玫瑰色": "Rose color", "将单空行": "刪除單行空白", @@ -2270,4 +2270,4 @@ "标注节点的行数范围": "標註節點的行數範圍", "默认 True": "默認 True", "将两个PDF拼接": "將兩個PDF拼接" -} \ No newline at end of file +} diff --git a/multi_language.py b/multi_language.py index a20fb5af..a807dbd0 100644 --- a/multi_language.py +++ b/multi_language.py @@ -182,12 +182,12 @@ cached_translation = read_map_from_json(language=LANG) def trans(word_to_translate, language, special=False): if len(word_to_translate) == 0: return {} from crazy_functions.crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency - from toolbox import get_conf, ChatBotWithCookies - 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') + from toolbox import get_conf, ChatBotWithCookies, load_chat_cookies + + cookies = load_chat_cookies() llm_kwargs = { - 'api_key': API_KEY, - 'llm_model': LLM_MODEL, + 'api_key': cookies['api_key'], + 'llm_model': cookies['llm_model'], 'top_p':1.0, 'max_length': None, 'temperature':0.4, @@ -245,15 +245,15 @@ def trans(word_to_translate, language, special=False): def trans_json(word_to_translate, language, special=False): if len(word_to_translate) == 0: return {} from crazy_functions.crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency - from toolbox import get_conf, ChatBotWithCookies - 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') + from toolbox import get_conf, ChatBotWithCookies, load_chat_cookies + + cookies = load_chat_cookies() llm_kwargs = { - 'api_key': API_KEY, - 'llm_model': LLM_MODEL, + 'api_key': cookies['api_key'], + 'llm_model': cookies['llm_model'], 'top_p':1.0, 'max_length': None, - 'temperature':0.1, + 'temperature':0.4, } import random N_EACH_REQ = random.randint(16, 32) diff --git a/request_llms/bridge_all.py b/request_llms/bridge_all.py index 8dece548..689b1f97 100644 --- a/request_llms/bridge_all.py +++ b/request_llms/bridge_all.py @@ -431,16 +431,48 @@ if "chatglm_onnx" in AVAIL_LLM_MODELS: }) except: print(trimmed_format_exc()) -if "qwen" in AVAIL_LLM_MODELS: +if "qwen-local" in AVAIL_LLM_MODELS: + try: + from .bridge_qwen_local import predict_no_ui_long_connection as qwen_local_noui + from .bridge_qwen_local import predict as qwen_local_ui + model_info.update({ + "qwen-local": { + "fn_with_ui": qwen_local_ui, + "fn_without_ui": qwen_local_noui, + "endpoint": None, + "max_token": 4096, + "tokenizer": tokenizer_gpt35, + "token_cnt": get_token_num_gpt35, + } + }) + except: + print(trimmed_format_exc()) +if "qwen-turbo" in AVAIL_LLM_MODELS or "qwen-plus" in AVAIL_LLM_MODELS or "qwen-max" in AVAIL_LLM_MODELS: # zhipuai try: from .bridge_qwen import predict_no_ui_long_connection as qwen_noui from .bridge_qwen import predict as qwen_ui model_info.update({ - "qwen": { + "qwen-turbo": { "fn_with_ui": qwen_ui, "fn_without_ui": qwen_noui, "endpoint": None, - "max_token": 4096, + "max_token": 6144, + "tokenizer": tokenizer_gpt35, + "token_cnt": get_token_num_gpt35, + }, + "qwen-plus": { + "fn_with_ui": qwen_ui, + "fn_without_ui": qwen_noui, + "endpoint": None, + "max_token": 30720, + "tokenizer": tokenizer_gpt35, + "token_cnt": get_token_num_gpt35, + }, + "qwen-max": { + "fn_with_ui": qwen_ui, + "fn_without_ui": qwen_noui, + "endpoint": None, + "max_token": 28672, "tokenizer": tokenizer_gpt35, "token_cnt": get_token_num_gpt35, } @@ -552,7 +584,7 @@ if "deepseekcoder" in AVAIL_LLM_MODELS: # deepseekcoder "fn_with_ui": deepseekcoder_ui, "fn_without_ui": deepseekcoder_noui, "endpoint": None, - "max_token": 4096, + "max_token": 2048, "tokenizer": tokenizer_gpt35, "token_cnt": get_token_num_gpt35, } diff --git a/request_llms/bridge_chatgpt.py b/request_llms/bridge_chatgpt.py index e55ad37a..faad6cfd 100644 --- a/request_llms/bridge_chatgpt.py +++ b/request_llms/bridge_chatgpt.py @@ -51,7 +51,8 @@ def decode_chunk(chunk): chunkjson = json.loads(chunk_decoded[6:]) has_choices = 'choices' in chunkjson if has_choices: choice_valid = (len(chunkjson['choices']) > 0) - if has_choices and choice_valid: has_content = "content" in chunkjson['choices'][0]["delta"] + if has_choices and choice_valid: has_content = ("content" in chunkjson['choices'][0]["delta"]) + if has_content: has_content = (chunkjson['choices'][0]["delta"]["content"] is not None) if has_choices and choice_valid: has_role = "role" in chunkjson['choices'][0]["delta"] except: pass @@ -101,20 +102,25 @@ def predict_no_ui_long_connection(inputs, llm_kwargs, history=[], sys_prompt="", result = '' json_data = None while True: - try: chunk = next(stream_response).decode() + try: chunk = next(stream_response) 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() + chunk = next(stream_response) # 失败了，重试一次？再失败就没办法了。 + chunk_decoded, chunkjson, has_choices, choice_valid, has_content, has_role = decode_chunk(chunk) + if len(chunk_decoded)==0: continue + if not chunk_decoded.startswith('data:'): + error_msg = get_full_error(chunk, 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) - if ('data: [DONE]' in chunk): break # api2d 正常完成 - json_data = json.loads(chunk.lstrip('data:'))['choices'][0] + if ('data: [DONE]' in chunk_decoded): break # api2d 正常完成 + # 提前读取一些信息 （用于判断异常） + if has_choices and not choice_valid: + # 一些垃圾第三方接口的出现这样的错误 + continue + json_data = chunkjson['choices'][0] delta = json_data["delta"] if len(delta) == 0: break if "role" in delta: continue diff --git a/request_llms/bridge_chatgpt_vision.py b/request_llms/bridge_chatgpt_vision.py index e84bc0b7..ebcf9689 100644 --- a/request_llms/bridge_chatgpt_vision.py +++ b/request_llms/bridge_chatgpt_vision.py @@ -15,29 +15,16 @@ import requests import base64 import os import glob +from toolbox import get_conf, update_ui, is_any_api_key, select_api_key, what_keys, clip_history, trimmed_format_exc, is_the_upload_folder, \ + update_ui_lastest_msg, get_max_token, encode_image, have_any_recent_upload_image_files + -from toolbox import get_conf, update_ui, is_any_api_key, select_api_key, what_keys, clip_history, trimmed_format_exc, is_the_upload_folder, update_ui_lastest_msg, get_max_token proxies, TIMEOUT_SECONDS, MAX_RETRY, API_ORG, AZURE_CFG_ARRAY = \ get_conf('proxies', 'TIMEOUT_SECONDS', 'MAX_RETRY', 'API_ORG', 'AZURE_CFG_ARRAY') 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 have_any_recent_upload_image_files(chatbot): - _5min = 5 * 60 - if chatbot is None: return False, None # chatbot is None - most_recent_uploaded = chatbot._cookies.get("most_recent_uploaded", None) - if not most_recent_uploaded: return False, None # most_recent_uploaded is None - if time.time() - most_recent_uploaded["time"] < _5min: - most_recent_uploaded = chatbot._cookies.get("most_recent_uploaded", None) - path = most_recent_uploaded['path'] - file_manifest = [f for f in glob.glob(f'{path}/**/*.jpg', recursive=True)] - file_manifest += [f for f in glob.glob(f'{path}/**/*.jpeg', recursive=True)] - file_manifest += [f for f in glob.glob(f'{path}/**/*.png', recursive=True)] - if len(file_manifest) == 0: return False, None - return True, file_manifest # most_recent_uploaded is new - else: - return False, None # most_recent_uploaded is too old def report_invalid_key(key): if get_conf("BLOCK_INVALID_APIKEY"): @@ -258,10 +245,6 @@ def handle_error(inputs, llm_kwargs, chatbot, history, chunk_decoded, error_msg, chatbot[-1] = (chatbot[-1][0], f"[Local Message] 异常 \n\n{tb_str} \n\n{regular_txt_to_markdown(chunk_decoded)}") return chatbot, history -# Function to encode the image -def encode_image(image_path): - with open(image_path, "rb") as image_file: - return base64.b64encode(image_file.read()).decode('utf-8') def generate_payload(inputs, llm_kwargs, history, system_prompt, image_paths): """ diff --git a/request_llms/bridge_deepseekcoder.py b/request_llms/bridge_deepseekcoder.py index 2242eec7..89964abe 100644 --- a/request_llms/bridge_deepseekcoder.py +++ b/request_llms/bridge_deepseekcoder.py @@ -6,6 +6,7 @@ from toolbox import ProxyNetworkActivate from toolbox import get_conf from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns from threading import Thread +import torch def download_huggingface_model(model_name, max_retry, local_dir): from huggingface_hub import snapshot_download @@ -36,9 +37,46 @@ class GetCoderLMHandle(LocalLLMHandle): # tokenizer = download_huggingface_model(model_name, max_retry=128, local_dir=local_dir) tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True) self._streamer = TextIteratorStreamer(tokenizer) - model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True) + device_map = { + "transformer.word_embeddings": 0, + "transformer.word_embeddings_layernorm": 0, + "lm_head": 0, + "transformer.h": 0, + "transformer.ln_f": 0, + "model.embed_tokens": 0, + "model.layers": 0, + "model.norm": 0, + } + + # 检查量化配置 + quantization_type = get_conf('LOCAL_MODEL_QUANT') + if get_conf('LOCAL_MODEL_DEVICE') != 'cpu': - model = model.cuda() + if quantization_type == "INT8": + from transformers import BitsAndBytesConfig + # 使用 INT8 量化 + model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, load_in_8bit=True, + device_map=device_map) + elif quantization_type == "INT4": + from transformers import BitsAndBytesConfig + # 使用 INT4 量化 + bnb_config = BitsAndBytesConfig( + load_in_4bit=True, + bnb_4bit_use_double_quant=True, + bnb_4bit_quant_type="nf4", + bnb_4bit_compute_dtype=torch.bfloat16 + ) + model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, + quantization_config=bnb_config, device_map=device_map) + else: + # 使用默认的 FP16 + model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, + torch_dtype=torch.bfloat16, device_map=device_map) + else: + # CPU 模式 + model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, + torch_dtype=torch.bfloat16) + return model, tokenizer def llm_stream_generator(self, **kwargs): @@ -54,7 +92,10 @@ class GetCoderLMHandle(LocalLLMHandle): query, max_length, top_p, temperature, history = adaptor(kwargs) history.append({ 'role': 'user', 'content': query}) messages = history - inputs = self._tokenizer.apply_chat_template(messages, return_tensors="pt").to(self._model.device) + inputs = self._tokenizer.apply_chat_template(messages, return_tensors="pt") + if inputs.shape[1] > max_length: + inputs = inputs[:, -max_length:] + inputs = inputs.to(self._model.device) generation_kwargs = dict( inputs=inputs, max_new_tokens=max_length, diff --git a/request_llms/bridge_qwen.py b/request_llms/bridge_qwen.py index 85a4d80c..18877b94 100644 --- a/request_llms/bridge_qwen.py +++ b/request_llms/bridge_qwen.py @@ -1,67 +1,62 @@ -model_name = "Qwen" -cmd_to_install = "`pip install -r request_llms/requirements_qwen.txt`" - - -from transformers import AutoModel, AutoTokenizer import time -import threading -import importlib -from toolbox import update_ui, get_conf, ProxyNetworkActivate -from multiprocessing import Process, Pipe -from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns +import os +from toolbox import update_ui, get_conf, update_ui_lastest_msg +from toolbox import check_packages, report_exception +model_name = 'Qwen' +def predict_no_ui_long_connection(inputs, llm_kwargs, history=[], sys_prompt="", observe_window=[], console_slience=False): + """ + ⭐多线程方法 + 函数的说明请见 request_llms/bridge_all.py + """ + watch_dog_patience = 5 + response = "" -# ------------------------------------------------------------------------------------------------------------------------ -# 🔌💻 Local Model -# ------------------------------------------------------------------------------------------------------------------------ -class GetQwenLMHandle(LocalLLMHandle): + from .com_qwenapi import QwenRequestInstance + sri = QwenRequestInstance() + for response in sri.generate(inputs, llm_kwargs, history, sys_prompt): + if len(observe_window) >= 1: + observe_window[0] = response + if len(observe_window) >= 2: + if (time.time()-observe_window[1]) > watch_dog_patience: raise RuntimeError("程序终止。") + return response - def load_model_info(self): - # 🏃‍♂️🏃‍♂️🏃‍♂️ 子进程执行 - self.model_name = model_name - self.cmd_to_install = cmd_to_install +def predict(inputs, llm_kwargs, plugin_kwargs, chatbot, history=[], system_prompt='', stream = True, additional_fn=None): + """ + ⭐单线程方法 + 函数的说明请见 request_llms/bridge_all.py + """ + chatbot.append((inputs, "")) + yield from update_ui(chatbot=chatbot, history=history) - def load_model_and_tokenizer(self): - # 🏃‍♂️🏃‍♂️🏃‍♂️ 子进程执行 - import os, glob - import os - import platform - from modelscope import AutoModelForCausalLM, AutoTokenizer, GenerationConfig + # 尝试导入依赖，如果缺少依赖，则给出安装建议 + try: + check_packages(["dashscope"]) + except: + yield from update_ui_lastest_msg(f"导入软件依赖失败。使用该模型需要额外依赖，安装方法```pip install --upgrade dashscope```。", + chatbot=chatbot, history=history, delay=0) + return - with ProxyNetworkActivate('Download_LLM'): - model_id = 'qwen/Qwen-7B-Chat' - self._tokenizer = AutoTokenizer.from_pretrained('Qwen/Qwen-7B-Chat', trust_remote_code=True, resume_download=True) - # use fp16 - model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True, fp16=True).eval() - model.generation_config = GenerationConfig.from_pretrained(model_id, trust_remote_code=True) # 可指定不同的生成长度、top_p等相关超参 - self._model = model + # 检查DASHSCOPE_API_KEY + if get_conf("DASHSCOPE_API_KEY") == "": + yield from update_ui_lastest_msg(f"请配置 DASHSCOPE_API_KEY。", + chatbot=chatbot, history=history, delay=0) + return - return self._model, self._tokenizer + if additional_fn is not None: + from core_functional import handle_core_functionality + inputs, history = handle_core_functionality(additional_fn, inputs, history, chatbot) - def llm_stream_generator(self, **kwargs): - # 🏃‍♂️🏃‍♂️🏃‍♂️ 子进程执行 - def adaptor(kwargs): - query = kwargs['query'] - max_length = kwargs['max_length'] - top_p = kwargs['top_p'] - temperature = kwargs['temperature'] - history = kwargs['history'] - return query, max_length, top_p, temperature, history + # 开始接收回复 + from .com_qwenapi import QwenRequestInstance + sri = QwenRequestInstance() + for response in sri.generate(inputs, llm_kwargs, history, system_prompt): + chatbot[-1] = (inputs, response) + yield from update_ui(chatbot=chatbot, history=history) - query, max_length, top_p, temperature, history = adaptor(kwargs) - - for response in self._model.chat(self._tokenizer, query, history=history, stream=True): - yield response - - def try_to_import_special_deps(self, **kwargs): - # import something that will raise error if the user does not install requirement_*.txt - # 🏃‍♂️🏃‍♂️🏃‍♂️ 主进程执行 - import importlib - importlib.import_module('modelscope') - - -# ------------------------------------------------------------------------------------------------------------------------ -# 🔌💻 GPT-Academic Interface -# ------------------------------------------------------------------------------------------------------------------------ -predict_no_ui_long_connection, predict = get_local_llm_predict_fns(GetQwenLMHandle, model_name) \ No newline at end of file + # 总结输出 + if response == f"[Local Message] 等待{model_name}响应中 ...": + response = f"[Local Message] {model_name}响应异常 ..." + history.extend([inputs, response]) + yield from update_ui(chatbot=chatbot, history=history) \ No newline at end of file diff --git a/request_llms/bridge_qwen_local.py b/request_llms/bridge_qwen_local.py new file mode 100644 index 00000000..e6c2dd5c --- /dev/null +++ b/request_llms/bridge_qwen_local.py @@ -0,0 +1,59 @@ +model_name = "Qwen_Local" +cmd_to_install = "`pip install -r request_llms/requirements_qwen_local.txt`" + +from toolbox import ProxyNetworkActivate, get_conf +from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns + + + +# ------------------------------------------------------------------------------------------------------------------------ +# 🔌💻 Local Model +# ------------------------------------------------------------------------------------------------------------------------ +class GetQwenLMHandle(LocalLLMHandle): + + def load_model_info(self): + # 🏃‍♂️🏃‍♂️🏃‍♂️ 子进程执行 + self.model_name = model_name + self.cmd_to_install = cmd_to_install + + def load_model_and_tokenizer(self): + # 🏃‍♂️🏃‍♂️🏃‍♂️ 子进程执行 + # from modelscope import AutoModelForCausalLM, AutoTokenizer, GenerationConfig + from transformers import AutoModelForCausalLM, AutoTokenizer + from transformers.generation import GenerationConfig + with ProxyNetworkActivate('Download_LLM'): + model_id = get_conf('QWEN_LOCAL_MODEL_SELECTION') + self._tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True, resume_download=True) + # use fp16 + model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True).eval() + model.generation_config = GenerationConfig.from_pretrained(model_id, trust_remote_code=True) # 可指定不同的生成长度、top_p等相关超参 + self._model = model + + return self._model, self._tokenizer + + def llm_stream_generator(self, **kwargs): + # 🏃‍♂️🏃‍♂️🏃‍♂️ 子进程执行 + def adaptor(kwargs): + query = kwargs['query'] + max_length = kwargs['max_length'] + top_p = kwargs['top_p'] + temperature = kwargs['temperature'] + history = kwargs['history'] + return query, max_length, top_p, temperature, history + + query, max_length, top_p, temperature, history = adaptor(kwargs) + + for response in self._model.chat_stream(self._tokenizer, query, history=history): + yield response + + def try_to_import_special_deps(self, **kwargs): + # import something that will raise error if the user does not install requirement_*.txt + # 🏃‍♂️🏃‍♂️🏃‍♂️ 主进程执行 + import importlib + importlib.import_module('modelscope') + + +# ------------------------------------------------------------------------------------------------------------------------ +# 🔌💻 GPT-Academic Interface +# ------------------------------------------------------------------------------------------------------------------------ +predict_no_ui_long_connection, predict = get_local_llm_predict_fns(GetQwenLMHandle, model_name) \ No newline at end of file diff --git a/request_llms/bridge_spark.py b/request_llms/bridge_spark.py index 6ba39ee7..1fe31ce9 100644 --- a/request_llms/bridge_spark.py +++ b/request_llms/bridge_spark.py @@ -26,7 +26,7 @@ def predict_no_ui_long_connection(inputs, llm_kwargs, history=[], sys_prompt="", from .com_sparkapi import SparkRequestInstance sri = SparkRequestInstance() - for response in sri.generate(inputs, llm_kwargs, history, sys_prompt): + for response in sri.generate(inputs, llm_kwargs, history, sys_prompt, use_image_api=False): if len(observe_window) >= 1: observe_window[0] = response if len(observe_window) >= 2: @@ -52,7 +52,7 @@ def predict(inputs, llm_kwargs, plugin_kwargs, chatbot, history=[], system_promp # 开始接收回复 from .com_sparkapi import SparkRequestInstance sri = SparkRequestInstance() - for response in sri.generate(inputs, llm_kwargs, history, system_prompt): + for response in sri.generate(inputs, llm_kwargs, history, system_prompt, use_image_api=True): chatbot[-1] = (inputs, response) yield from update_ui(chatbot=chatbot, history=history) diff --git a/request_llms/com_qwenapi.py b/request_llms/com_qwenapi.py new file mode 100644 index 00000000..5807600d --- /dev/null +++ b/request_llms/com_qwenapi.py @@ -0,0 +1,94 @@ +from http import HTTPStatus +from toolbox import get_conf +import threading +import logging + +timeout_bot_msg = '[Local Message] Request timeout. Network error.' + +class QwenRequestInstance(): + def __init__(self): + import dashscope + self.time_to_yield_event = threading.Event() + self.time_to_exit_event = threading.Event() + self.result_buf = "" + + def validate_key(): + DASHSCOPE_API_KEY = get_conf("DASHSCOPE_API_KEY") + if DASHSCOPE_API_KEY == '': return False + return True + + if not validate_key(): + raise RuntimeError('请配置 DASHSCOPE_API_KEY') + dashscope.api_key = get_conf("DASHSCOPE_API_KEY") + + + def generate(self, inputs, llm_kwargs, history, system_prompt): + # import _thread as thread + from dashscope import Generation + QWEN_MODEL = { + 'qwen-turbo': Generation.Models.qwen_turbo, + 'qwen-plus': Generation.Models.qwen_plus, + 'qwen-max': Generation.Models.qwen_max, + }[llm_kwargs['llm_model']] + top_p = llm_kwargs.get('top_p', 0.8) + if top_p == 0: top_p += 1e-5 + if top_p == 1: top_p -= 1e-5 + + self.result_buf = "" + responses = Generation.call( + model=QWEN_MODEL, + messages=generate_message_payload(inputs, llm_kwargs, history, system_prompt), + top_p=top_p, + temperature=llm_kwargs.get('temperature', 1.0), + result_format='message', + stream=True, + incremental_output=True + ) + + for response in responses: + if response.status_code == HTTPStatus.OK: + if response.output.choices[0].finish_reason == 'stop': + yield self.result_buf + break + elif response.output.choices[0].finish_reason == 'length': + self.result_buf += "[Local Message] 生成长度过长，后续输出被截断" + yield self.result_buf + break + else: + self.result_buf += response.output.choices[0].message.content + yield self.result_buf + else: + self.result_buf += f"[Local Message] 请求错误：状态码：{response.status_code}，错误码:{response.code}，消息：{response.message}" + yield self.result_buf + break + logging.info(f'[raw_input] {inputs}') + logging.info(f'[response] {self.result_buf}') + return self.result_buf + + +def generate_message_payload(inputs, llm_kwargs, history, system_prompt): + conversation_cnt = len(history) // 2 + if system_prompt == '': system_prompt = 'Hello!' + messages = [{"role": "user", "content": system_prompt}, {"role": "assistant", "content": "Certainly!"}] + 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) + return messages diff --git a/request_llms/com_sparkapi.py b/request_llms/com_sparkapi.py index 5c1a3a44..3f667c18 100644 --- a/request_llms/com_sparkapi.py +++ b/request_llms/com_sparkapi.py @@ -1,4 +1,4 @@ -from toolbox import get_conf +from toolbox import get_conf, get_pictures_list, encode_image import base64 import datetime import hashlib @@ -65,18 +65,19 @@ class SparkRequestInstance(): self.gpt_url = "ws://spark-api.xf-yun.com/v1.1/chat" self.gpt_url_v2 = "ws://spark-api.xf-yun.com/v2.1/chat" self.gpt_url_v3 = "ws://spark-api.xf-yun.com/v3.1/chat" + self.gpt_url_img = "wss://spark-api.cn-huabei-1.xf-yun.com/v2.1/image" self.time_to_yield_event = threading.Event() self.time_to_exit_event = threading.Event() self.result_buf = "" - def generate(self, inputs, llm_kwargs, history, system_prompt): + def generate(self, inputs, llm_kwargs, history, system_prompt, use_image_api=False): llm_kwargs = llm_kwargs history = history system_prompt = system_prompt import _thread as thread - thread.start_new_thread(self.create_blocking_request, (inputs, llm_kwargs, history, system_prompt)) + thread.start_new_thread(self.create_blocking_request, (inputs, llm_kwargs, history, system_prompt, use_image_api)) while True: self.time_to_yield_event.wait(timeout=1) if self.time_to_yield_event.is_set(): @@ -85,14 +86,20 @@ class SparkRequestInstance(): return self.result_buf - def create_blocking_request(self, inputs, llm_kwargs, history, system_prompt): + def create_blocking_request(self, inputs, llm_kwargs, history, system_prompt, use_image_api): if llm_kwargs['llm_model'] == 'sparkv2': gpt_url = self.gpt_url_v2 elif llm_kwargs['llm_model'] == 'sparkv3': gpt_url = self.gpt_url_v3 else: gpt_url = self.gpt_url - + file_manifest = [] + if use_image_api and llm_kwargs.get('most_recent_uploaded'): + if llm_kwargs['most_recent_uploaded'].get('path'): + file_manifest = get_pictures_list(llm_kwargs['most_recent_uploaded']['path']) + if len(file_manifest) > 0: + print('正在使用讯飞图片理解API') + gpt_url = self.gpt_url_img wsParam = Ws_Param(self.appid, self.api_key, self.api_secret, gpt_url) websocket.enableTrace(False) wsUrl = wsParam.create_url() @@ -101,9 +108,8 @@ class SparkRequestInstance(): def on_open(ws): import _thread as thread thread.start_new_thread(run, (ws,)) - def run(ws, *args): - data = json.dumps(gen_params(ws.appid, *ws.all_args)) + data = json.dumps(gen_params(ws.appid, *ws.all_args, file_manifest)) ws.send(data) # 收到websocket消息的处理 @@ -142,9 +148,18 @@ class SparkRequestInstance(): ws.all_args = (inputs, llm_kwargs, history, system_prompt) ws.run_forever(sslopt={"cert_reqs": ssl.CERT_NONE}) -def generate_message_payload(inputs, llm_kwargs, history, system_prompt): +def generate_message_payload(inputs, llm_kwargs, history, system_prompt, file_manifest): conversation_cnt = len(history) // 2 - messages = [{"role": "system", "content": system_prompt}] + messages = [] + if file_manifest: + base64_images = [] + for image_path in file_manifest: + base64_images.append(encode_image(image_path)) + for img_s in base64_images: + if img_s not in str(messages): + messages.append({"role": "user", "content": img_s, "content_type": "image"}) + else: + messages = [{"role": "system", "content": system_prompt}] if conversation_cnt: for index in range(0, 2*conversation_cnt, 2): what_i_have_asked = {} @@ -167,7 +182,7 @@ def generate_message_payload(inputs, llm_kwargs, history, system_prompt): return messages -def gen_params(appid, inputs, llm_kwargs, history, system_prompt): +def gen_params(appid, inputs, llm_kwargs, history, system_prompt, file_manifest): """ 通过appid和用户的提问来生成请参数 """ @@ -176,6 +191,8 @@ def gen_params(appid, inputs, llm_kwargs, history, system_prompt): "sparkv2": "generalv2", "sparkv3": "generalv3", } + domains_select = domains[llm_kwargs['llm_model']] + if file_manifest: domains_select = 'image' data = { "header": { "app_id": appid, @@ -183,7 +200,7 @@ def gen_params(appid, inputs, llm_kwargs, history, system_prompt): }, "parameter": { "chat": { - "domain": domains[llm_kwargs['llm_model']], + "domain": domains_select, "temperature": llm_kwargs["temperature"], "random_threshold": 0.5, "max_tokens": 4096, @@ -192,7 +209,7 @@ def gen_params(appid, inputs, llm_kwargs, history, system_prompt): }, "payload": { "message": { - "text": generate_message_payload(inputs, llm_kwargs, history, system_prompt) + "text": generate_message_payload(inputs, llm_kwargs, history, system_prompt, file_manifest) } } } diff --git a/request_llms/local_llm_class.py b/request_llms/local_llm_class.py index 413df03f..ec7cfd21 100644 --- a/request_llms/local_llm_class.py +++ b/request_llms/local_llm_class.py @@ -183,11 +183,11 @@ class LocalLLMHandle(Process): def stream_chat(self, **kwargs): # ⭐run in main process if self.get_state() == "`准备就绪`": - yield "`正在等待线程锁，排队中请稍后 ...`" + yield "`正在等待线程锁，排队中请稍候 ...`" with self.threadLock: if self.parent.poll(): - yield "`排队中请稍后 ...`" + yield "`排队中请稍候 ...`" self.clear_pending_messages() self.parent.send(kwargs) std_out = "" diff --git a/request_llms/requirements_chatglm_onnx.txt b/request_llms/requirements_chatglm_onnx.txt index 54811472..2cd11f69 100644 --- a/request_llms/requirements_chatglm_onnx.txt +++ b/request_llms/requirements_chatglm_onnx.txt @@ -6,5 +6,3 @@ sentencepiece numpy onnxruntime sentencepiece -streamlit -streamlit-chat diff --git a/request_llms/requirements_moss.txt b/request_llms/requirements_moss.txt index c27907c2..544b25f0 100644 --- a/request_llms/requirements_moss.txt +++ b/request_llms/requirements_moss.txt @@ -5,5 +5,4 @@ accelerate matplotlib huggingface_hub triton -streamlit diff --git a/request_llms/requirements_qwen.txt b/request_llms/requirements_qwen.txt index 3d7d62a0..5899464f 100644 --- a/request_llms/requirements_qwen.txt +++ b/request_llms/requirements_qwen.txt @@ -1,2 +1 @@ -modelscope -transformers_stream_generator \ No newline at end of file +dashscope \ No newline at end of file diff --git a/request_llms/requirements_qwen_local.txt b/request_llms/requirements_qwen_local.txt new file mode 100644 index 00000000..de6bf3c9 --- /dev/null +++ b/request_llms/requirements_qwen_local.txt @@ -0,0 +1,5 @@ +modelscope +transformers_stream_generator +auto-gptq +optimum +urllib3<2 \ No newline at end of file diff --git a/requirements.txt b/requirements.txt index fa78f6dd..52e851de 100644 --- a/requirements.txt +++ b/requirements.txt @@ -2,6 +2,7 @@ pydantic==1.10.11 pypdf2==2.12.1 tiktoken>=0.3.3 requests[socks] +protobuf==3.18 transformers>=4.27.1 scipdf_parser>=0.52 python-markdown-math diff --git a/tests/test_llms.py b/tests/test_llms.py index 8b685972..347c6b99 100644 --- a/tests/test_llms.py +++ b/tests/test_llms.py @@ -16,8 +16,9 @@ if __name__ == "__main__": # from request_llms.bridge_jittorllms_llama import predict_no_ui_long_connection # from request_llms.bridge_claude import predict_no_ui_long_connection # from request_llms.bridge_internlm import predict_no_ui_long_connection - from request_llms.bridge_deepseekcoder import predict_no_ui_long_connection - # from request_llms.bridge_qwen import predict_no_ui_long_connection + # from request_llms.bridge_deepseekcoder import predict_no_ui_long_connection + # from request_llms.bridge_qwen_7B import predict_no_ui_long_connection + from request_llms.bridge_qwen_local import predict_no_ui_long_connection # from request_llms.bridge_spark import predict_no_ui_long_connection # from request_llms.bridge_zhipu import predict_no_ui_long_connection # from request_llms.bridge_chatglm3 import predict_no_ui_long_connection diff --git a/tests/test_plugins.py b/tests/test_plugins.py index 8470895d..13ec259e 100644 --- a/tests/test_plugins.py +++ b/tests/test_plugins.py @@ -48,11 +48,11 @@ if __name__ == "__main__": # for lang in ["English", "French", "Japanese", "Korean", "Russian", "Italian", "German", "Portuguese", "Arabic"]: # plugin_test(plugin='crazy_functions.批量Markdown翻译->Markdown翻译指定语言', main_input="README.md", advanced_arg={"advanced_arg": lang}) - # plugin_test(plugin='crazy_functions.Langchain知识库->知识库问答', main_input="./") + # plugin_test(plugin='crazy_functions.知识库文件注入->知识库文件注入', main_input="./") - # plugin_test(plugin='crazy_functions.Langchain知识库->读取知识库作答', main_input="What is the installation method？") + # plugin_test(plugin='crazy_functions.知识库文件注入->读取知识库作答', main_input="What is the installation method？") - # plugin_test(plugin='crazy_functions.Langchain知识库->读取知识库作答', main_input="远程云服务器部署？") + # plugin_test(plugin='crazy_functions.知识库文件注入->读取知识库作答', main_input="远程云服务器部署？") # plugin_test(plugin='crazy_functions.Latex输出PDF结果->Latex翻译中文并重新编译PDF', main_input="2210.03629") diff --git a/tests/test_utils.py b/tests/test_utils.py index 1fdca1eb..c87908f6 100644 --- a/tests/test_utils.py +++ b/tests/test_utils.py @@ -56,11 +56,11 @@ vt.get_plugin_handle = silence_stdout_fn(get_plugin_handle) vt.get_plugin_default_kwargs = silence_stdout_fn(get_plugin_default_kwargs) vt.get_chat_handle = silence_stdout_fn(get_chat_handle) vt.get_chat_default_kwargs = silence_stdout_fn(get_chat_default_kwargs) -vt.chat_to_markdown_str = chat_to_markdown_str +vt.chat_to_markdown_str = (chat_to_markdown_str) proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION, CHATBOT_HEIGHT, LAYOUT, API_KEY = \ vt.get_conf('proxies', 'WEB_PORT', 'LLM_MODEL', 'CONCURRENT_COUNT', 'AUTHENTICATION', 'CHATBOT_HEIGHT', 'LAYOUT', 'API_KEY') -def plugin_test(main_input, plugin, advanced_arg=None): +def plugin_test(main_input, plugin, advanced_arg=None, debug=True): from rich.live import Live from rich.markdown import Markdown @@ -72,7 +72,10 @@ def plugin_test(main_input, plugin, advanced_arg=None): plugin_kwargs['main_input'] = main_input if advanced_arg is not None: plugin_kwargs['plugin_kwargs'] = advanced_arg - my_working_plugin = silence_stdout(plugin)(**plugin_kwargs) + if debug: + my_working_plugin = (plugin)(**plugin_kwargs) + else: + my_working_plugin = silence_stdout(plugin)(**plugin_kwargs) with Live(Markdown(""), auto_refresh=False, vertical_overflow="visible") as live: for cookies, chat, hist, msg in my_working_plugin: diff --git a/tests/test_vector_plugins.py b/tests/test_vector_plugins.py new file mode 100644 index 00000000..9b75463b --- /dev/null +++ b/tests/test_vector_plugins.py @@ -0,0 +1,17 @@ +""" +对项目中的各个插件进行测试。运行方法：直接运行 python tests/test_plugins.py +""" + + +import os, sys +def validate_path(): dir_name = os.path.dirname(__file__); root_dir_assume = os.path.abspath(dir_name + '/..'); os.chdir(root_dir_assume); sys.path.append(root_dir_assume) +validate_path() # 返回项目根路径 + +if __name__ == "__main__": + from tests.test_utils import plugin_test + + plugin_test(plugin='crazy_functions.知识库问答->知识库文件注入', main_input="./README.md") + + plugin_test(plugin='crazy_functions.知识库问答->读取知识库作答', main_input="What is the installation method？") + + plugin_test(plugin='crazy_functions.知识库问答->读取知识库作答', main_input="远程云服务器部署？") \ No newline at end of file diff --git a/themes/common.js b/themes/common.js index 849cb9a5..8743b3ca 100644 --- a/themes/common.js +++ b/themes/common.js @@ -1,9 +1,13 @@ +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= +// 第 1 部分: 工具函数 +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= + function gradioApp() { // https://github.com/GaiZhenbiao/ChuanhuChatGPT/tree/main/web_assets/javascript const elems = document.getElementsByTagName('gradio-app'); const elem = elems.length == 0 ? document : elems[0]; if (elem !== document) { - elem.getElementById = function(id) { + elem.getElementById = function (id) { return document.getElementById(id); }; } @@ -12,31 +16,76 @@ function gradioApp() { function setCookie(name, value, days) { var expires = ""; - + if (days) { - var date = new Date(); - date.setTime(date.getTime() + (days * 24 * 60 * 60 * 1000)); - expires = "; expires=" + date.toUTCString(); + var date = new Date(); + date.setTime(date.getTime() + (days * 24 * 60 * 60 * 1000)); + expires = "; expires=" + date.toUTCString(); } - + document.cookie = name + "=" + value + expires + "; path=/"; } function getCookie(name) { var decodedCookie = decodeURIComponent(document.cookie); var cookies = decodedCookie.split(';'); - + for (var i = 0; i < cookies.length; i++) { - var cookie = cookies[i].trim(); - - if (cookie.indexOf(name + "=") === 0) { - return cookie.substring(name.length + 1, cookie.length); - } + var cookie = cookies[i].trim(); + + if (cookie.indexOf(name + "=") === 0) { + return cookie.substring(name.length + 1, cookie.length); + } } - + return null; - } - +} + +let toastCount = 0; +function toast_push(msg, duration) { + duration = isNaN(duration) ? 3000 : duration; + const existingToasts = document.querySelectorAll('.toast'); + existingToasts.forEach(toast => { + toast.style.top = `${parseInt(toast.style.top, 10) - 70}px`; + }); + const m = document.createElement('div'); + m.innerHTML = msg; + m.classList.add('toast'); + m.style.cssText = `font-size: var(--text-md) !important; color: rgb(255, 255, 255); background-color: rgba(0, 0, 0, 0.6); padding: 10px 15px; border-radius: 4px; position: fixed; top: ${50 + toastCount * 70}%; left: 50%; transform: translateX(-50%); width: auto; text-align: center; transition: top 0.3s;`; + document.body.appendChild(m); + setTimeout(function () { + m.style.opacity = '0'; + setTimeout(function () { + document.body.removeChild(m); + toastCount--; + }, 500); + }, duration); + toastCount++; +} + +function toast_up(msg) { + var m = document.getElementById('toast_up'); + if (m) { + document.body.removeChild(m); // remove the loader from the body + } + m = document.createElement('div'); + m.id = 'toast_up'; + m.innerHTML = msg; + m.style.cssText = "font-size: var(--text-md) !important; color: rgb(255, 255, 255); background-color: rgba(0, 0, 100, 0.6); padding: 10px 15px; margin: 0 0 0 -60px; border-radius: 4px; position: fixed; top: 50%; left: 50%; width: auto; text-align: center;"; + document.body.appendChild(m); +} +function toast_down() { + var m = document.getElementById('toast_up'); + if (m) { + document.body.removeChild(m); // remove the loader from the body + } +} + + +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= +// 第 2 部分: 复制按钮 +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= + function addCopyButton(botElement) { // https://github.com/GaiZhenbiao/ChuanhuChatGPT/tree/main/web_assets/javascript // Copy bot button @@ -49,7 +98,7 @@ function addCopyButton(botElement) { // messageBtnColumnElement.remove(); return; } - + var copyButton = document.createElement('button'); copyButton.classList.add('copy-bot-btn'); copyButton.setAttribute('aria-label', 'Copy'); @@ -98,47 +147,61 @@ function chatbotContentChanged(attempt = 1, force = false) { } } -function chatbotAutoHeight(){ + + +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= +// 第 3 部分: chatbot动态高度调整 +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= + +function chatbotAutoHeight() { // 自动调整高度 - function update_height(){ - var { panel_height_target, chatbot_height, chatbot } = get_elements(true); - if (panel_height_target!=chatbot_height) - { - var pixelString = panel_height_target.toString() + 'px'; - chatbot.style.maxHeight = pixelString; chatbot.style.height = pixelString; + function update_height() { + var { height_target, chatbot_height, chatbot } = get_elements(true); + if (height_target != chatbot_height) { + var pixelString = height_target.toString() + 'px'; + chatbot.style.maxHeight = pixelString; chatbot.style.height = pixelString; } } - function update_height_slow(){ - var { panel_height_target, chatbot_height, chatbot } = get_elements(); - if (panel_height_target!=chatbot_height) - { - new_panel_height = (panel_height_target - chatbot_height)*0.5 + chatbot_height; - if (Math.abs(new_panel_height - panel_height_target) < 10){ - new_panel_height = panel_height_target; + function update_height_slow() { + var { height_target, chatbot_height, chatbot } = get_elements(); + if (height_target != chatbot_height) { + new_panel_height = (height_target - chatbot_height) * 0.5 + chatbot_height; + if (Math.abs(new_panel_height - height_target) < 10) { + new_panel_height = height_target; } - // console.log(chatbot_height, panel_height_target, new_panel_height); + // console.log(chatbot_height, height_target, new_panel_height); var pixelString = new_panel_height.toString() + 'px'; - chatbot.style.maxHeight = pixelString; chatbot.style.height = pixelString; + chatbot.style.maxHeight = pixelString; chatbot.style.height = pixelString; } } - + monitoring_input_box() update_height(); - setInterval(function() { + setInterval(function () { update_height_slow() - }, 50); // 每100毫秒执行一次 + }, 50); // 每50毫秒执行一次 } -function GptAcademicJavaScriptInit(LAYOUT = "LEFT-RIGHT") { - chatbotIndicator = gradioApp().querySelector('#gpt-chatbot > div.wrap'); - var chatbotObserver = new MutationObserver(() => { - chatbotContentChanged(1); - }); - chatbotObserver.observe(chatbotIndicator, { attributes: true, childList: true, subtree: true }); - if (LAYOUT === "LEFT-RIGHT") {chatbotAutoHeight();} +swapped = false; +function swap_input_area() { + // Get the elements to be swapped + var element1 = document.querySelector("#input-panel"); + var element2 = document.querySelector("#basic-panel"); + + // Get the parent of the elements + var parent = element1.parentNode; + + // Get the next sibling of element2 + var nextSibling = element2.nextSibling; + + // Swap the elements + parent.insertBefore(element2, element1); + parent.insertBefore(element1, nextSibling); + if (swapped) {swapped = false;} + else {swapped = true;} } -function get_elements(consider_state_panel=false) { +function get_elements(consider_state_panel = false) { var chatbot = document.querySelector('#gpt-chatbot > div.wrap.svelte-18telvq'); if (!chatbot) { chatbot = document.querySelector('#gpt-chatbot'); @@ -147,17 +210,292 @@ function get_elements(consider_state_panel=false) { const panel2 = document.querySelector('#basic-panel').getBoundingClientRect() const panel3 = document.querySelector('#plugin-panel').getBoundingClientRect(); // const panel4 = document.querySelector('#interact-panel').getBoundingClientRect(); - const panel5 = document.querySelector('#input-panel2').getBoundingClientRect(); const panel_active = document.querySelector('#state-panel').getBoundingClientRect(); - if (consider_state_panel || panel_active.height < 25){ + if (consider_state_panel || panel_active.height < 25) { document.state_panel_height = panel_active.height; } // 25 是chatbot的label高度, 16 是右侧的gap - var panel_height_target = panel1.height + panel2.height + panel3.height + 0 + 0 - 25 + 16*2; + var height_target = panel1.height + panel2.height + panel3.height + 0 + 0 - 25 + 16 * 2; // 禁止动态的state-panel高度影响 - panel_height_target = panel_height_target + (document.state_panel_height-panel_active.height) - var panel_height_target = parseInt(panel_height_target); + height_target = height_target + (document.state_panel_height - panel_active.height) + var height_target = parseInt(height_target); var chatbot_height = chatbot.style.height; + // 交换输入区位置，使得输入区始终可用 + if (!swapped){ + if (panel1.top!=0 && panel1.top < 0){ swap_input_area(); } + } + else if (swapped){ + if (panel2.top!=0 && panel2.top > 0){ swap_input_area(); } + } + // 调整高度 + const err_tor = 5; + if (Math.abs(panel1.left - chatbot.getBoundingClientRect().left) < err_tor){ + // 是否处于窄屏模式 + height_target = window.innerHeight * 0.6; + }else{ + // 调整高度 + const chatbot_height_exceed = 15; + const chatbot_height_exceed_m = 10; + b_panel = Math.max(panel1.bottom, panel2.bottom, panel3.bottom) + if (b_panel >= window.innerHeight - chatbot_height_exceed) { + height_target = window.innerHeight - chatbot.getBoundingClientRect().top - chatbot_height_exceed_m; + } + else if (b_panel < window.innerHeight * 0.75) { + height_target = window.innerHeight * 0.8; + } + } var chatbot_height = parseInt(chatbot_height); - return { panel_height_target, chatbot_height, chatbot }; + return { height_target, chatbot_height, chatbot }; +} + + + +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= +// 第 4 部分: 粘贴、拖拽文件上传 +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= + +var elem_upload = null; +var elem_upload_float = null; +var elem_input_main = null; +var elem_input_float = null; +var elem_chatbot = null; +var exist_file_msg = '⚠️请先删除上传区（左上方）中的历史文件，再尝试上传。' + +function add_func_paste(input) { + let paste_files = []; + if (input) { + input.addEventListener("paste", async function (e) { + const clipboardData = e.clipboardData || window.clipboardData; + const items = clipboardData.items; + if (items) { + for (i = 0; i < items.length; i++) { + if (items[i].kind === "file") { // 确保是文件类型 + const file = items[i].getAsFile(); + // 将每一个粘贴的文件添加到files数组中 + paste_files.push(file); + e.preventDefault(); // 避免粘贴文件名到输入框 + } + } + if (paste_files.length > 0) { + // 按照文件列表执行批量上传逻辑 + await upload_files(paste_files); + paste_files = [] + + } + } + }); + } +} + +function add_func_drag(elem) { + if (elem) { + const dragEvents = ["dragover"]; + const leaveEvents = ["dragleave", "dragend", "drop"]; + + const onDrag = function (e) { + e.preventDefault(); + e.stopPropagation(); + if (elem_upload_float.querySelector("input[type=file]")) { + toast_up('⚠️释放以上传文件') + } else { + toast_up(exist_file_msg) + } + }; + + const onLeave = function (e) { + toast_down(); + e.preventDefault(); + e.stopPropagation(); + }; + + dragEvents.forEach(event => { + elem.addEventListener(event, onDrag); + }); + + leaveEvents.forEach(event => { + elem.addEventListener(event, onLeave); + }); + + elem.addEventListener("drop", async function (e) { + const files = e.dataTransfer.files; + await upload_files(files); + }); + } +} + +async function upload_files(files) { + const uploadInputElement = elem_upload_float.querySelector("input[type=file]"); + let totalSizeMb = 0 + if (files && files.length > 0) { + // 执行具体的上传逻辑 + if (uploadInputElement) { + for (let i = 0; i < files.length; i++) { + // 将从文件数组中获取的文件大小(单位为字节)转换为MB， + totalSizeMb += files[i].size / 1024 / 1024; + } + // 检查文件总大小是否超过20MB + if (totalSizeMb > 20) { + toast_push('⚠️文件夹大于 20MB 🚀上传文件中', 3000) + // return; // 如果超过了指定大小, 可以不进行后续上传操作 + } + // 监听change事件， 原生Gradio可以实现 + // uploadInputElement.addEventListener('change', function(){replace_input_string()}); + let event = new Event("change"); + Object.defineProperty(event, "target", { value: uploadInputElement, enumerable: true }); + Object.defineProperty(event, "currentTarget", { value: uploadInputElement, enumerable: true }); + Object.defineProperty(uploadInputElement, "files", { value: files, enumerable: true }); + uploadInputElement.dispatchEvent(event); + } else { + toast_push(exist_file_msg, 3000) + } + } +} + +function begin_loading_status() { + // Create the loader div and add styling + var loader = document.createElement('div'); + loader.id = 'Js_File_Loading'; + loader.style.position = "absolute"; + loader.style.top = "50%"; + loader.style.left = "50%"; + loader.style.width = "60px"; + loader.style.height = "60px"; + loader.style.border = "16px solid #f3f3f3"; + loader.style.borderTop = "16px solid #3498db"; + loader.style.borderRadius = "50%"; + loader.style.animation = "spin 2s linear infinite"; + loader.style.transform = "translate(-50%, -50%)"; + document.body.appendChild(loader); // Add the loader to the body + // Set the CSS animation keyframes + var styleSheet = document.createElement('style'); + // styleSheet.type = 'text/css'; + styleSheet.id = 'Js_File_Loading_Style' + styleSheet.innerText = ` + @keyframes spin { + 0% { transform: rotate(0deg); } + 100% { transform: rotate(360deg); } + }`; + document.head.appendChild(styleSheet); +} + +function cancel_loading_status() { + var loadingElement = document.getElementById('Js_File_Loading'); + if (loadingElement) { + document.body.removeChild(loadingElement); // remove the loader from the body + } + var loadingStyle = document.getElementById('Js_File_Loading_Style'); + if (loadingStyle) { + document.head.removeChild(loadingStyle); + } + let clearButton = document.querySelectorAll('div[id*="elem_upload"] button[aria-label="Clear"]'); + for (let button of clearButton) { + button.addEventListener('click', function () { + setTimeout(function () { + register_upload_event(); + }, 50); + }); + } +} + +function register_upload_event() { + elem_upload_float = document.getElementById('elem_upload_float') + const upload_component = elem_upload_float.querySelector("input[type=file]"); + if (upload_component) { + upload_component.addEventListener('change', function (event) { + toast_push('正在上传中，请稍等。', 2000); + begin_loading_status(); + }); + } +} + +function monitoring_input_box() { + register_upload_event(); + + elem_upload = document.getElementById('elem_upload') + elem_upload_float = document.getElementById('elem_upload_float') + elem_input_main = document.getElementById('user_input_main') + elem_input_float = document.getElementById('user_input_float') + elem_chatbot = document.getElementById('gpt-chatbot') + + if (elem_input_main) { + if (elem_input_main.querySelector("textarea")) { + add_func_paste(elem_input_main.querySelector("textarea")) + } + } + if (elem_input_float) { + if (elem_input_float.querySelector("textarea")) { + add_func_paste(elem_input_float.querySelector("textarea")) + } + } + if (elem_chatbot) { + add_func_drag(elem_chatbot) + } +} + + +// 监视页面变化 +window.addEventListener("DOMContentLoaded", function () { + // const ga = document.getElementsByTagName("gradio-app"); + gradioApp().addEventListener("render", monitoring_input_box); +}); + + + + + +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= +// 第 5 部分: 音频按钮样式变化 +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= + +function audio_fn_init() { + let audio_component = document.getElementById('elem_audio'); + if (audio_component) { + let buttonElement = audio_component.querySelector('button'); + let specificElement = audio_component.querySelector('.hide.sr-only'); + specificElement.remove(); + + buttonElement.childNodes[1].nodeValue = '启动麦克风'; + buttonElement.addEventListener('click', function (event) { + event.stopPropagation(); + toast_push('您启动了麦克风!下一步请点击“实时语音对话”启动语音对话。'); + }); + + // 查找语音插件按钮 + let buttons = document.querySelectorAll('button'); + let audio_button = null; + for (let button of buttons) { + if (button.textContent.includes('语音')) { + audio_button = button; + break; + } + } + if (audio_button) { + audio_button.addEventListener('click', function () { + toast_push('您点击了“实时语音对话”启动语音对话。'); + }); + let parent_element = audio_component.parentElement; // 将buttonElement移动到audio_button的内部 + audio_button.appendChild(audio_component); + buttonElement.style.cssText = 'border-color: #00ffe0;border-width: 2px; height: 25px;' + parent_element.remove(); + audio_component.style.cssText = 'width: 250px;right: 0px;display: inline-flex;flex-flow: row-reverse wrap;place-content: stretch space-between;align-items: center;background-color: #ffffff00;'; + } + + } +} + + + + +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= +// 第 6 部分: JS初始化函数 +// -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= + +function GptAcademicJavaScriptInit(LAYOUT = "LEFT-RIGHT") { + audio_fn_init(); + chatbotIndicator = gradioApp().querySelector('#gpt-chatbot > div.wrap'); + var chatbotObserver = new MutationObserver(() => { + chatbotContentChanged(1); + }); + chatbotObserver.observe(chatbotIndicator, { attributes: true, childList: true, subtree: true }); + if (LAYOUT === "LEFT-RIGHT") { chatbotAutoHeight(); } } \ No newline at end of file diff --git a/themes/cookies.py b/themes/cookies.py new file mode 100644 index 00000000..e69de29b diff --git a/themes/green.css b/themes/green.css index dd109d53..870af482 100644 --- a/themes/green.css +++ b/themes/green.css @@ -256,13 +256,13 @@ textarea.svelte-1pie7s6 { max-height: 95% !important; overflow-y: auto !important; }*/ -.app.svelte-1mya07g.svelte-1mya07g { +/* .app.svelte-1mya07g.svelte-1mya07g { max-width: 100%; position: relative; padding: var(--size-4); width: 100%; height: 100%; -} +} */ .gradio-container-3-32-2 h1 { font-weight: 700 !important; diff --git a/themes/theme.py b/themes/theme.py index f59db9f8..5664f737 100644 --- a/themes/theme.py +++ b/themes/theme.py @@ -1,6 +1,14 @@ -import gradio as gr +import pickle +import base64 +import uuid from toolbox import get_conf -THEME = get_conf('THEME') + +""" +-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- +第 1 部分 +加载主题相关的工具函数 +-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- +""" def load_dynamic_theme(THEME): adjust_dynamic_theme = None @@ -20,4 +28,91 @@ def load_dynamic_theme(THEME): theme_declaration = "" return adjust_theme, advanced_css, theme_declaration, adjust_dynamic_theme -adjust_theme, advanced_css, theme_declaration, _ = load_dynamic_theme(THEME) \ No newline at end of file +adjust_theme, advanced_css, theme_declaration, _ = load_dynamic_theme(get_conf('THEME')) + + + + + + +""" +-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- +第 2 部分 +cookie相关工具函数 +-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- +""" + +def init_cookie(cookies, chatbot): + # 为每一位访问的用户赋予一个独一无二的uuid编码 + cookies.update({'uuid': uuid.uuid4()}) + return cookies + +def to_cookie_str(d): + # Pickle the dictionary and encode it as a string + pickled_dict = pickle.dumps(d) + cookie_value = base64.b64encode(pickled_dict).decode('utf-8') + return cookie_value + +def from_cookie_str(c): + # Decode the base64-encoded string and unpickle it into a dictionary + pickled_dict = base64.b64decode(c.encode('utf-8')) + return pickle.loads(pickled_dict) + + + + + +""" +-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- +第 3 部分 +内嵌的javascript代码 +-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- +""" + +js_code_for_css_changing = """(css) => { + var existingStyles = document.querySelectorAll("body > gradio-app > div > style") + for (var i = 0; i < existingStyles.length; i++) { + var style = existingStyles[i]; + style.parentNode.removeChild(style); + } + var existingStyles = document.querySelectorAll("style[data-loaded-css]"); + for (var i = 0; i < existingStyles.length; i++) { + var style = existingStyles[i]; + style.parentNode.removeChild(style); + } + var styleElement = document.createElement('style'); + styleElement.setAttribute('data-loaded-css', 'placeholder'); + styleElement.innerHTML = css; + document.body.appendChild(styleElement); +} +""" + +js_code_for_darkmode_init = """(dark) => { + dark = dark == "True"; + if (document.querySelectorAll('.dark').length) { + if (!dark){ + document.querySelectorAll('.dark').forEach(el => el.classList.remove('dark')); + } + } else { + if (dark){ + document.querySelector('body').classList.add('dark'); + } + } +} +""" + +js_code_for_toggle_darkmode = """() => { + if (document.querySelectorAll('.dark').length) { + document.querySelectorAll('.dark').forEach(el => el.classList.remove('dark')); + } else { + document.querySelector('body').classList.add('dark'); + } +}""" + + +js_code_for_persistent_cookie_init = """(persistent_cookie) => { + return getCookie("persistent_cookie"); +} +""" + + diff --git a/toolbox.py b/toolbox.py index 8747afdb..154b54ca 100644 --- a/toolbox.py +++ b/toolbox.py @@ -4,6 +4,7 @@ import time import inspect import re import os +import base64 import gradio import shutil import glob @@ -79,6 +80,7 @@ def ArgsGeneralWrapper(f): 'max_length': max_length, 'temperature':temperature, 'client_ip': request.client.host, + 'most_recent_uploaded': cookies.get('most_recent_uploaded') } plugin_kwargs = { "advanced_arg": plugin_advanced_arg, @@ -178,12 +180,15 @@ def HotReload(f): 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 + if get_conf('PLUGIN_HOT_RELOAD'): + @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 + else: + return f """ @@ -561,7 +566,8 @@ def promote_file_to_downloadzone(file, rename_file=None, chatbot=None): user_name = get_user(chatbot) else: user_name = default_user_name - + if not os.path.exists(file): + raise FileNotFoundError(f'文件{file}不存在') user_path = get_log_folder(user_name, plugin_name=None) if file_already_in_downloadzone(file, user_path): new_path = file @@ -577,7 +583,8 @@ def promote_file_to_downloadzone(file, rename_file=None, chatbot=None): if chatbot is not None: if 'files_to_promote' in chatbot._cookies: current = chatbot._cookies['files_to_promote'] else: current = [] - chatbot._cookies.update({'files_to_promote': [new_path] + current}) + if new_path not in current: # 避免把同一个文件添加多次 + chatbot._cookies.update({'files_to_promote': [new_path] + current}) return new_path @@ -602,6 +609,64 @@ def del_outdated_uploads(outdate_time_seconds, target_path_base=None): except: pass return + +def html_local_file(file): + base_path = os.path.dirname(__file__) # 项目目录 + if os.path.exists(str(file)): + file = f'file={file.replace(base_path, ".")}' + return file + + +def html_local_img(__file, layout='left', max_width=None, max_height=None, md=True): + style = '' + if max_width is not None: + style += f"max-width: {max_width};" + if max_height is not None: + style += f"max-height: {max_height};" + __file = html_local_file(__file) + a = f'
' + if md: + a = f'![{__file}]({__file})' + return a + +def file_manifest_filter_type(file_list, filter_: list = None): + new_list = [] + if not filter_: filter_ = ['png', 'jpg', 'jpeg'] + for file in file_list: + if str(os.path.basename(file)).split('.')[-1] in filter_: + new_list.append(html_local_img(file, md=False)) + else: + new_list.append(file) + return new_list + +def to_markdown_tabs(head: list, tabs: list, alignment=':---:', column=False): + """ + Args: + head: 表头：[] + tabs: 表值：[[列1], [列2], [列3], [列4]] + alignment: :--- 左对齐， :---: 居中对齐， ---: 右对齐 + column: True to keep data in columns, False to keep data in rows (default). + Returns: + A string representation of the markdown table. + """ + if column: + transposed_tabs = list(map(list, zip(*tabs))) + else: + transposed_tabs = tabs + # Find the maximum length among the columns + max_len = max(len(column) for column in transposed_tabs) + + tab_format = "| %s " + tabs_list = "".join([tab_format % i for i in head]) + '|\n' + tabs_list += "".join([tab_format % alignment for i in head]) + '|\n' + + for i in range(max_len): + row_data = [tab[i] if i < len(tab) else '' for tab in transposed_tabs] + row_data = file_manifest_filter_type(row_data, filter_=None) + tabs_list += "".join([tab_format % i for i in row_data]) + '|\n' + + return tabs_list + def on_file_uploaded(request: gradio.Request, files, chatbot, txt, txt2, checkboxes, cookies): """ 当文件被上传时的回调函数 @@ -626,16 +691,15 @@ def on_file_uploaded(request: gradio.Request, files, chatbot, txt, txt2, checkbo this_file_path = pj(target_path_base, file_origin_name) shutil.move(file.name, this_file_path) upload_msg += extract_archive(file_path=this_file_path, dest_dir=this_file_path+'.extract') - - # 整理文件集合 - moved_files = [fp for fp in glob.glob(f'{target_path_base}/**/*', recursive=True)] + if "浮动输入区" in checkboxes: txt, txt2 = "", target_path_base else: txt, txt2 = target_path_base, "" - # 输出消息 - moved_files_str = '\t\n\n'.join(moved_files) + # 整理文件集合 输出消息 + moved_files = [fp for fp in glob.glob(f'{target_path_base}/**/*', recursive=True)] + moved_files_str = to_markdown_tabs(head=['文件'], tabs=[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}' + @@ -856,7 +920,14 @@ def read_single_conf_with_lru_cache(arg): @lru_cache(maxsize=128) 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 + """ + 本项目的所有配置都集中在config.py中。 修改配置有三种方法，您只需要选择其中一种即可： + - 直接修改config.py + - 创建并修改config_private.py + - 修改环境变量（修改docker-compose.yml等价于修改容器内部的环境变量） + + 注意：如果您使用docker-compose部署，请修改docker-compose（等价于修改容器内部的环境变量） + """ res = [] for arg in args: r = read_single_conf_with_lru_cache(arg) @@ -937,14 +1008,19 @@ def clip_history(inputs, history, tokenizer, max_token_limit): def get_token_num(txt): return len(tokenizer.encode(txt, disallowed_special=())) input_token_num = get_token_num(inputs) + + if max_token_limit < 5000: output_token_expect = 256 # 4k & 2k models + elif max_token_limit < 9000: output_token_expect = 512 # 8k models + else: output_token_expect = 1024 # 16k & 32k models + if input_token_num < max_token_limit * 3 / 4: # 当输入部分的token占比小于限制的3/4时，裁剪时 # 1. 把input的余量留出来 max_token_limit = max_token_limit - input_token_num # 2. 把输出用的余量留出来 - max_token_limit = max_token_limit - 128 + max_token_limit = max_token_limit - output_token_expect # 3. 如果余量太小了，直接清除历史 - if max_token_limit < 128: + if max_token_limit < output_token_expect: history = [] return history else: @@ -1053,7 +1129,7 @@ def get_user(chatbotwithcookies): class ProxyNetworkActivate(): """ - 这段代码定义了一个名为TempProxy的空上下文管理器, 用于给一小段代码上代理 + 这段代码定义了一个名为ProxyNetworkActivate的空上下文管理器, 用于给一小段代码上代理 """ def __init__(self, task=None) -> None: self.task = task @@ -1198,6 +1274,35 @@ def get_chat_default_kwargs(): return default_chat_kwargs + +def get_pictures_list(path): + file_manifest = [f for f in glob.glob(f'{path}/**/*.jpg', recursive=True)] + file_manifest += [f for f in glob.glob(f'{path}/**/*.jpeg', recursive=True)] + file_manifest += [f for f in glob.glob(f'{path}/**/*.png', recursive=True)] + return file_manifest + + +def have_any_recent_upload_image_files(chatbot): + _5min = 5 * 60 + if chatbot is None: return False, None # chatbot is None + most_recent_uploaded = chatbot._cookies.get("most_recent_uploaded", None) + if not most_recent_uploaded: return False, None # most_recent_uploaded is None + if time.time() - most_recent_uploaded["time"] < _5min: + most_recent_uploaded = chatbot._cookies.get("most_recent_uploaded", None) + path = most_recent_uploaded['path'] + file_manifest = get_pictures_list(path) + if len(file_manifest) == 0: return False, None + return True, file_manifest # most_recent_uploaded is new + else: + return False, None # most_recent_uploaded is too old + + +# Function to encode the image +def encode_image(image_path): + with open(image_path, "rb") as image_file: + return base64.b64encode(image_file.read()).decode('utf-8') + + def get_max_token(llm_kwargs): from request_llms.bridge_all import model_info return model_info[llm_kwargs['llm_model']]['max_token'] diff --git a/version b/version index 5f6de09c..62e6aa5c 100644 --- a/version +++ b/version @@ -1,5 +1,5 @@ { - "version": 3.61, + "version": 3.64, "show_feature": true, - "new_feature": "修复潜在的多用户冲突问题 <-> 接入Deepseek Coder <-> AutoGen多智能体插件测试版 <-> 修复本地模型在Windows下的加载BUG <-> 支持文心一言v4和星火v3 <-> 支持GLM3和智谱的API <-> 解决本地模型并发BUG <-> 支持动态追加基础功能按钮" + "new_feature": "支持直接拖拽文件到上传区 <-> 支持将图片粘贴到输入区 <-> 修复若干隐蔽的内存BUG <-> 修复多用户冲突问题 <-> 接入Deepseek Coder <-> AutoGen多智能体插件测试版" }