124 lines
3.4 KiB
Markdown
124 lines
3.4 KiB
Markdown
|
|
|
|
<div align="center">
|
|
> 简体中文 < | <a href="https://www.google.com/search?q=best+website+to+learn+chinese&newwindow=1&sca_esv=a76695392a9980a7&rlz=1C1CHBD_zh-HKHK1072HK1072&sxsrf=ADLYWIKib09skMzGw8JqpTv2AWB6Xk8uZQ%3A1716986674152&ei=MiNXZsP3COrk2roPt-SrgA0&oq=best+website+to+learn+chin&gs_lp=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-YAwCIBgGQBgq6BgYIARABGBSSBwoxLjE2LjUuMS4xoAf2rwE&sclient=gws-wiz-serp" target="_blank">English</a>
|
|
</div>
|
|
|
|
## 结构
|
|
|
|
每一个服务都放在 ./node 中，根据各自的 readme 分别配置。
|
|
|
|
你需要准备的环境变量：
|
|
|
|
```bash
|
|
export DEEPSEEK_API_TOKEN=xxx
|
|
export OMCP_DISCORD_SERVER_IP=xxx
|
|
export OMCP_DISCORD_SERVER_PORT=xxx
|
|
export OMCP_DISCORD_TOKEN=xxx
|
|
```
|
|
|
|
|
|
---
|
|
|
|
## 架构
|
|
|
|
```mermaid
|
|
graph TB
|
|
core(Lagrage.Core)
|
|
onebot(Lagrange.onebot)
|
|
vecdb(vecdb)
|
|
llm(LLM)
|
|
intent(intent tree)
|
|
|
|
core(Lagrange.Core) --> onebot(Lagrange.onebot)
|
|
|
|
onebot -->|query| intent
|
|
intent -->|intent| onebot
|
|
|
|
subgraph Intent Recognition
|
|
intent -->|query| vecdb
|
|
vecdb -->|ktop| intent
|
|
intent -->|ktop,query| llm
|
|
llm -->|intent| intent
|
|
end
|
|
|
|
subgraph execution
|
|
onebot --> command{intent}
|
|
command --> query
|
|
command --> upload
|
|
command --> ...
|
|
end
|
|
|
|
subgraph third party
|
|
LLM
|
|
Google
|
|
server
|
|
end
|
|
|
|
query --> LLM
|
|
query --> Google
|
|
upload --> server
|
|
```
|
|
|
|
- `Lagrange.onebot` --> 📁bot
|
|
- `vecdb` --> 📁rag
|
|
- `intent tree` --> 📁prompt
|
|
|
|
---
|
|
|
|
## 接口规范
|
|
|
|
http 接口满足 `HttpResponse` 所示。
|
|
|
|
```typescript
|
|
interface HttpResponse<T> {
|
|
code: number,
|
|
data: CommonResponse<T>
|
|
}
|
|
|
|
interface CommonResponse<T> {
|
|
code: number,
|
|
data?: T,
|
|
msg?: string
|
|
}
|
|
```
|
|
|
|
---
|
|
|
|
## 开发须知
|
|
|
|
- 非必要，请不要随意宣传本项目。
|
|
- 虽然曾经无数个 QQ 相关的项目都死了，但是基本的 API 端口算是传承了下来。拉格朗日的返回类型，请参考 [go-cqhttp 帮助中心 - API 篇](https://docs.go-cqhttp.org/api/) 中的内容。
|
|
|
|
---
|
|
|
|
## 启动
|
|
|
|
```bash
|
|
# 1. 启动 拉格朗日
|
|
tsc
|
|
pm2 start dist/main.js --name Lagrange.onebot
|
|
pm2 start rag/main.py --name rag
|
|
```
|
|
|
|
---
|
|
|
|
## 测试命令
|
|
|
|
### 重训练 embedding -> intent 分类层
|
|
|
|
```bash
|
|
curl -X POST http://127.0.0.1:8081/intent/retrain-embedding-mapping
|
|
```
|
|
|
|
### 获取意图
|
|
|
|
```bash
|
|
curl -X POST -H "Content-Type: application/json" -d '{"query": "真的开线程是要tcl指令去改的"}' http://127.0.0.1:8081/intent/get-intent-recogition
|
|
```
|
|
|
|
### 获取向量数据库中的 topk
|
|
|
|
```bash
|
|
curl -X POST -H "Content-Type: application/json" -d '{"query": "这个插件有什么比较好的文档吗？"}' http://127.0.0.1:8081/vecdb/similarity_search_with_score
|
|
``` |