Dify源码学习-工作流的角度

原创已于 2024-09-07 09:09:17 修改 · 2.2k 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#学习 #python #人工智能

于 2024-09-06 23:58:51 首次发布

菜鸟小白学习分享，如有不对欢迎指正

下载安装

Dify 依赖以下工具和库：

Docker
Docker Compose
Node.js v18.x (LTS)
npm version 8.x.x or Yarn
Python version 3.10.x

注：需要严格按照官网文档环境信息

具体安装步骤可从github上复制粘贴，傻瓜式操作，but 首先你需要创建一个虚拟环境 https://github.com/langgenius/dify/blob/main/CONTRIBUTING_CN.md

0.6.12版本后使用poetry管理包

ps：不要怪dify只支持上传图片，而且前端传参还是base64格式，因为模型接口就只支持处理image，编码还必须是base64 文件上传随后再写吧

后端

Dify 的后端使用 Python 编写，使用 Flask 框架。它使用 SQLAlchemy 作为 ORM，使用 Celery 作为任务队列。授权逻辑通过 Flask-login 进行处理。

by the way: celery 初步理解为消费者模型，发送消息后，worker起一个‘进程’处理消息，（个人猜想是进程）个人对异步消息这块不是很了解，不再赘述

poetry run python -m celery -A app.celery worker -P gevent -c 1 --loglevel INFO -Q dataset,generation,mail,ops_trace,app_deletion

数据库

普通工作流调用大模型时，仅使用到了postgres, 猜想weaviate向量数据库只有对应查询数据库的时候使用

工作流对应DB中workflows这张表，其中节点信息储存在graph字段中，以下为字段信息,是一段JSON

{"nodes": [{"id": "1725434172434", "type": "custom", "data": {"type": "start", "title": "\u5f00\u59cb", "desc": "", "variables": [], "selected": false}, "position": {"x": 80, "y": 282}, "targetPosition": "left", "sourcePosition": "right", "positionAbsolute": {"x": 80, "y": 282}, "width": 244, "height": 54}, {"id": "llm", "type": "custom", "data": {"type": "llm", "title": "LLM", "desc": "", "variables": [], "model": {"provider": "ollama", "name": "llava:13b", "mode": "chat", "completion_params": {"temperature": 0.7}}, "prompt_template": [{"role": "system", "text": "\u5c06\u56fe\u7247\u4e2d\u4fe1\u606f\u63d0\u53d6\u51fa\u6765", "id": "3253d6ae-8331-424f-9d85-7c9ebff6b5f4"}], "context": {"enabled": false, "variable_selector": []}, "vision": {"enabled": false}, "memory": {"window": {"enabled": false, "size": 10}, "role_prefix": {"user": "", "assistant": ""}}, "selected": true}, "position": {"x": 379.03416092544296, "y": 282}, "targetPosition": "left", "sourcePosition": "right", "positionAbsolute": {"x": 379.03416092544296, "y": 282}, "width": 244, "height": 98, "selected": true}, {"id": "answer", "type": "custom", "data": {"type": "answer", "title": "\u76f4\u63a5\u56de\u590d", "desc": "", "variables": [], "answer": "{{#llm.text#}}", "selected": false}, "position": {"x": 680, "y": 282}, "targetPosition": "left", "sourcePosition": "right", "positionAbsolute": {"x": 680, "y": 282}, "width": 244, "height": 107}], "edges": [{"id": "1725434172434-llm", "source": "1725434172434", "sourceHandle": "source", "target": "llm", "targetHandle": "target", "type": "custom", "data": {"sourceType": "start", "targetType": "llm"}}, {"id": "llm-answer", "source": "llm", "sourceHandle": "source", "target": "answer", "targetHandle": "target", "type": "custom", "data": {"sourceType": "llm", "targetType": "answer"}}], "viewport": {"x": 94.87333799526743, "y": -74.36439111513661, "zoom": 1.035369168987668}}

其中nodes表示当前工作流共有几个节点，edges中有相关节点的引用关系