mirror of
https://github.com/alibaba/higress.git
synced 2026-04-20 19:47:29 +08:00
203 lines
7.2 KiB
Markdown
203 lines
7.2 KiB
Markdown
## 简介
|
||
|
||
**Note**
|
||
|
||
> 需要数据面的proxy wasm版本大于等于0.2.100
|
||
>
|
||
|
||
> 编译时,需要带上版本的tag,例如:tinygo build -o main.wasm -scheduler=none -target=wasi -gc=custom -tags="custommalloc nottinygc_finalizer proxy_wasm_version_0_2_100" ./
|
||
|
||
|
||
LLM响应结构化插件,用于根据默认或用户配置的Json Schema对AI的响应进行结构化,以便后续插件处理。注意目前只支持 `非流式响应`。
|
||
|
||
|
||
### 配置说明
|
||
|
||
| Name | Type | Requirement | Default | **Description** |
|
||
| --- | --- | --- | --- | --- |
|
||
| serviceName | str | required | - | AI服务或支持AI-Proxy的网关服务名称 |
|
||
| serviceDomain | str | optional | - | AI服务或支持AI-Proxy的网关服务域名/IP地址 |
|
||
| servicePath | str | optional | '/v1/chat/completions' | AI服务或支持AI-Proxy的网关服务基础路径 |
|
||
| serviceUrl | str | optional | - | AI服务或支持 AI-Proxy 的网关服务URL, 插件将自动提取Domain 和 Path, 用于填充未配置的 serviceDomain 或 servicePath |
|
||
| servicePort | int | optional | 443 | 网关服务端口 |
|
||
| serviceTimeout | int | optional | 50000 | 默认请求超时时间 |
|
||
| maxRetry | int | optional | 3 | 若回答无法正确提取格式化时重试次数 |
|
||
| contentPath | str | optional | "choices.0.message.content” | 从LLM回答中提取响应结果的gpath路径 |
|
||
| jsonSchema | str (json) | optional | - | 验证请求所参照的 jsonSchema, 为空只验证并返回合法Json格式响应 |
|
||
| enableSwagger | bool | optional | false | 是否启用 Swagger 协议进行验证 |
|
||
| enableOas3 | bool | optional | true | 是否启用 Oas3 协议进行验证 |
|
||
| enableContentDisposition | bool | optional | true | 是否启用 Content-Disposition 头部, 若启用则会在响应头中添加 `Content-Disposition: attachment; filename="response.json"` |
|
||
|
||
> 出于性能考虑,默认支持的最大 Json Schema 深度为 6。超过此深度的 Json Schema 将不用于验证响应,插件只会检查返回的响应是否为合法的 Json 格式。
|
||
|
||
|
||
### 请求和返回参数说明
|
||
|
||
- **请求参数**: 本插件请求格式为openai请求格式,包含`model`和`messages`字段,其中`model`为AI模型名称,`messages`为对话消息列表,每个消息包含`role`和`content`字段,`role`为消息角色,`content`为消息内容。
|
||
```json
|
||
{
|
||
"model": "gpt-4",
|
||
"messages": [
|
||
{"role": "user", "content": "give me a api doc for add the variable x to x+5"}
|
||
]
|
||
}
|
||
```
|
||
其他请求参数需参考配置的ai服务或网关服务的相应文档。
|
||
- **返回参数**:
|
||
- 返回满足定义的Json Schema约束的 `Json格式响应`
|
||
- 若未定义Json Schema,则返回合法的`Json格式响应`
|
||
- 若出现内部错误,则返回 `{ "Code": 10XX, "Msg": "错误信息提示" }`。
|
||
|
||
## 请求示例
|
||
|
||
```bash
|
||
curl -X POST "http://localhost:8001/v1/chat/completions" \
|
||
-H "Content-Type: application/json" \
|
||
-d '{
|
||
"model": "gpt-4",
|
||
"messages": [
|
||
{"role": "user", "content": "give me a api doc for add the variable x to x+5"}
|
||
]
|
||
}'
|
||
|
||
```
|
||
|
||
## 返回示例
|
||
### 正常返回
|
||
在正常情况下,系统应返回经过 JSON Schema 验证的 JSON 数据。如果未配置 JSON Schema,系统将返回符合 JSON 标准的合法 JSON 数据。
|
||
```json
|
||
{
|
||
"apiVersion": "1.0",
|
||
"request": {
|
||
"endpoint": "/add_to_five",
|
||
"method": "POST",
|
||
"port": 8080,
|
||
"headers": {
|
||
"Content-Type": "application/json"
|
||
},
|
||
"body": {
|
||
"x": 7
|
||
}
|
||
}
|
||
}
|
||
```
|
||
|
||
### 异常返回
|
||
在发生错误时,返回状态码为 `500`,返回内容为 JSON 格式的错误信息。包含错误码 `Code` 和错误信息 `Msg` 两个字段。
|
||
```json
|
||
{
|
||
"Code": 1006,
|
||
"Msg": "retry count exceed max retry count"
|
||
}
|
||
```
|
||
|
||
### 错误码说明
|
||
| 错误码 | 说明 |
|
||
| --- | --- |
|
||
| 1001 | 配置的Json Schema不是合法Json格式|
|
||
| 1002 | 配置的Json Schema编译失败,不是合法的Json Schema 格式或深度超出 jsonSchemaMaxDepth 且 rejectOnDepthExceeded 为true|
|
||
| 1003 | 无法在响应中提取合法的Json|
|
||
| 1004 | 响应为空字符串|
|
||
| 1005 | 响应不符合Json Schema定义|
|
||
| 1006 | 重试次数超过最大限制|
|
||
| 1007 | 无法获取响应内容,可能是上游服务配置错误或获取内容的ContentPath路径错误|
|
||
| 1008 | serciveDomain为空, 请注意serviceDomian或serviceUrl不能同时为空|
|
||
|
||
## 服务配置说明
|
||
本插件需要配置上游服务来支持出现异常时的自动重试机制, 支持的配置主要包括`支持openai接口的AI服务`或`本地网关服务`
|
||
|
||
### 支持openai接口的AI服务
|
||
以qwen为例,基本配置如下:
|
||
|
||
Yaml格式配置如下
|
||
```yaml
|
||
serviceName: qwen
|
||
serviceDomain: dashscope.aliyuncs.com
|
||
apiKey: [Your API Key]
|
||
servicePath: /compatible-mode/v1/chat/completions
|
||
jsonSchema:
|
||
title: ReasoningSchema
|
||
type: object
|
||
properties:
|
||
reasoning_steps:
|
||
type: array
|
||
items:
|
||
type: string
|
||
description: The reasoning steps leading to the final conclusion.
|
||
answer:
|
||
type: string
|
||
description: The final answer, taking into account the reasoning steps.
|
||
required:
|
||
- reasoning_steps
|
||
- answer
|
||
additionalProperties: false
|
||
```
|
||
|
||
JSON 格式配置
|
||
```json
|
||
{
|
||
"serviceName": "qwen",
|
||
"serviceUrl": "https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions",
|
||
"apiKey": "[Your API Key]",
|
||
"jsonSchema": {
|
||
"title": "ActionItemsSchema",
|
||
"type": "object",
|
||
"properties": {
|
||
"action_items": {
|
||
"type": "array",
|
||
"items": {
|
||
"type": "object",
|
||
"properties": {
|
||
"description": {
|
||
"type": "string",
|
||
"description": "Description of the action item."
|
||
},
|
||
"due_date": {
|
||
"type": ["string", "null"],
|
||
"description": "Due date for the action item, can be null if not specified."
|
||
},
|
||
"owner": {
|
||
"type": ["string", "null"],
|
||
"description": "Owner responsible for the action item, can be null if not specified."
|
||
}
|
||
},
|
||
"required": ["description", "due_date", "owner"],
|
||
"additionalProperties": false
|
||
},
|
||
"description": "List of action items from the meeting."
|
||
}
|
||
},
|
||
"required": ["action_items"],
|
||
"additionalProperties": false
|
||
}
|
||
}
|
||
```
|
||
|
||
### 本地网关服务
|
||
为了能复用已经配置好的服务,本插件也支持配置本地网关服务。例如,若网关已经配置好了[AI-proxy服务](../ai-proxy/README.md),则可以直接配置如下:
|
||
1. 创建一个固定IP为127.0.0.1的服务,例如localservice.static
|
||
```yaml
|
||
- name: outbound|10000||localservice.static
|
||
connect_timeout: 30s
|
||
type: LOGICAL_DNS
|
||
dns_lookup_family: V4_ONLY
|
||
lb_policy: ROUND_ROBIN
|
||
load_assignment:
|
||
cluster_name: outbound|8001||localservice.static
|
||
endpoints:
|
||
- lb_endpoints:
|
||
- endpoint:
|
||
address:
|
||
socket_address:
|
||
address: 127.0.0.1
|
||
port_value: 10000
|
||
```
|
||
2. 配置文件中添加localservice.static的服务配置
|
||
```yaml
|
||
serviceName: localservice
|
||
serviceDomain: 127.0.0.1
|
||
servicePort: 10000
|
||
```
|
||
3. 自动提取请求的Path,Header等信息
|
||
插件会自动提取请求的Path,Header等信息,从而避免对AI服务的重复配置。
|