一、前言
二、AI 代理
应用架构
provider:type: qwenapiTokens:- sk-xxxxxxxxxxxxxxxxxxxxxxtimeout: 1200000modelMapping:'gpt-3.5-turbo': qwen-turbo'gpt-4': qwen-max'*': qwen-max
三、AI 可观测
官方文档：https://help.aliyun.com/zh/mse/user-guide/ai-observable?spm=a2c4g.11186623.0.0.42b4d95f14d39z
enable: true
四、AI 内容安全
官方文档：https://help.aliyun.com/zh/mse/user-guide/ai-content-security?spm=a2c4g.11186623.0.0.4408488dYGAm2X
配置 AI 内容安全插件后，应用架构如下图所示：
serviceSource: dnsserviceName: green-cipservicePort: 443domain: green-cip.cn-hangzhou.aliyuncs.comak: xxxxxxxxxxxxxxxxxsk: xxxxxxxxxxxxxxxxx
五、AI Token 限流
官方文档：https://help.aliyun.com/zh/mse/user-guide/ai-token-current-limiting?spm=a2c4g.11186623.0.i4
创建一个 redis 服务并且在网关进行配置：
rule_name: default_rulerule_items:- limit_by_per_ip: from-remote-addrlimit_keys:- key: 0.0.0.0/0token_per_minute: 100redis:service_name: redis.staticservice_port: 6379username: xxxxxxpassword: xxxxxxrejected_code: 429rejected_msg: 您的请求频率过高，请稍后再试。
六、AI 缓存
官方文档：https://help.aliyun.com/zh/mse/user-guide/ai-cache?spm=a2c4g.11186623.0.0.3cdf306bbJSWdn
redis:serviceName: redis.staticservicePort: 6379timeout: 2000username: xxxxxx password: xxxxxx
七、AI Token 限流
官方文档：https://help.aliyun.com/zh/mse/user-guide/ai-rag?spm=a2c4g.11186623.0.0.24c77267ImcfGS
dashscope:apiKey: sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxserviceName: qwenservicePort: 443domain: dashscope.aliyuncs.comdashvector:apiKey: sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxserviceName: dashvectorservicePort: 443domain: vrs-cn-xxxxxxxxxxxxxx.dashvector.cn-hangzhou.aliyuncs.comcollection: xxxxxxxxxxxxxx
八、其他
-
prompt 模板[3]
-
prompt 装饰器[4]
templates:- name: "developer-chat"template:model: gpt-3.5-turbomessages:- role: systemcontent: "你是一个 {{program}} 专家, 你平时使用的编程语言为 {{language}}"- role: user content: "帮我写一个 {{program}} 程序, 你的返回结果里面应该只包含python代码"
请求 body 示例如下：
{"template": "developer-chat","properties": {"program": "冒泡排序","language": "python"}}
Prompt 装饰器允许用户在网关定义对 prompt 的修改操作，包括在原始请求之前和之后插入 message，配置示例如下，请求 body 与 openai 的请求一致。
prepend:- role: systemcontent: "请使用英语回答问题."append:- role: usercontent: "每次回答完问题，尝试进行反问"
response: enable: trueprompt: "帮我修改以下HTTP应答信息，要求：1. content-type修改为application/json；2. body由xml转化为json；3. 移除content-length。"provider: serviceName: qwendomain: dashscope.aliyuncs.com apiKey: sk-xxxxxxxxxxxxxxxxxxxxxxxxxxx
<?xml version='1.0' encoding='us-ascii'?><!--A SAMPLE set of slides--><slideshowtitle="Sample Slide Show"date="Date of publication"author="Yours Truly"><!-- TITLE SLIDE --><slide type="all"><title>Wake up to WonderWidgets!</title></slide><!-- OVERVIEW --><slide type="all"><title>Overview</title><item>Why <em>WonderWidgets</em> are great</item><item/><item>Who <em>buys</em> WonderWidgets</item></slide></slideshow>使用以上配置，通过网关访问 httpbin 的 /xml 接口，结果为：
{"slideshow": {"title": "Sample Slide Show","date": "Date of publication","author": "Yours Truly","slides": [{"type": "all","title": "Wake up to WonderWidgets!"},{"type": "all","title": "Overview","items": ["Why <em>WonderWidgets</em> are great","","Who <em>buys</em> WonderWidgets"]}]}}
