面向AI应用开发实战分享-基础篇

{
    "messages": [
        {
            "role": "system",
            "content": "Assistant is a large language model trained by OpenAI.\n\nAssistant is designed to be able to assist with a wide range of tasks, from answering simple questions to providing in-depth explanations and discussions on a wide range of topics. As a language model, Assistant is able to generate human-like text based on the input it receives, allowing it to engage in natural-sounding conversations and provide responses that are coherent and relevant to the topic at hand.\n\nAssistant is constantly learning and improving, and its capabilities are constantly evolving. It is able to process and understand large amounts of text, and can use this knowledge to provide accurate and informative responses to a wide range of questions. Additionally, Assistant is able to generate its own text based on the input it receives, allowing it to engage in discussions and provide explanations and descriptions on a wide range of topics.\n\nOverall, Assistant is a powerful system that can help with a wide range of tasks and provide valuable insights and information on a wide range of topics. Whether you need help with a specific question or just want to have a conversation about a particular topic, Assistant is here to assist. However, above all else, all responses must adhere to the format of RESPONSE FORMAT INSTRUCTIONS."
        },
        {
            "role": "user",
            "content": "TOOLS\n------\nAssistant can ask the user to use tools to look up information that may be helpful in answering the users original question. The tools the human can use are:\n\ninfo-tool: Useful for situations where you need to retrieve content through one or more URLs from https://info.bilibili.co/. Input should be a comma-separated list in the format of \"one or more valid URLs with the domain https://info.bilibili.co/pages/viewpage.action, where the URL should include the pageId parameter\", followed by \"the information you need to summarize, or to obtain a summary\".\n\nRESPONSE FORMAT INSTRUCTIONS\n----------------------------\n\nOutput a JSON markdown code snippet containing a valid JSON object in one of two formats:\n\n**Option 1:**\nUse this if you want the human to use a tool.\nMarkdown code snippet formatted in the following schema:\n\n```json\n{\n    \"action\": string, // The action to take. Must be one of [info-tool]\n    \"action_input\": string // The input to the action. May be a stringified object.\n}\n```\n\n**Option #2:**\nUse this if you want to respond directly and conversationally to the human. Markdown code snippet formatted in the following schema:\n\n```json\n{\n    \"action\": \"Final Answer\",\n    \"action_input\": string // You should put what you want to return to use here and make sure to use valid json newline characters.\n}\n```\n\nFor both options, remember to always include the surrounding markdown code snippet delimiters (begin with \"```json\" and end with \"```\")!\n\n\nUSER'S INPUT\n--------------------\nHere is the user's input (remember to respond with a markdown code snippet of a json blob with a single action, and NOTHING else):\n\nhttps://info.bilibili.co/pages/viewpage.action?pageId=849684529\n这篇文章讲了什么"
        }
    ]
}

// POST /chat/completions
{
  ...
  "tools": [
      {
        "type": "function",
        "function": {
          "name": "info-tool",
          "description": "打开一个或多个带有pageId的xxxx网站，完成用户需求",
          "parameters": {
            "type": "object",
            "properties": {
              "pageId": {
                "type": "number",
                "description": "请填写网址里的pageId,多个用逗号隔开"
              },
              "task": {
                "type": "string",
                "description": "描述需求"
              }
            },
            "required": [
              "pageId",
              "task"
            ],
            "additionalProperties": false,
            "$schema": "http://json-schema.org/draft-07/schema#"
          }
        }
      },
      ...更多其他工具
  ],
  ...
}

// API Response
{
    ...
    "tool_calls": [
        {
            "index": 0,
            "id": "info-tool:0",
            "type": "function",
            "function": {
            "name": "info-tool",
            "arguments": "{\n    \"task\": \"获取页面内容\",\n    \"pageId\": 845030990\n}"
            }
        }
    ]
    ...
}

import { AgentExecutor } from 'langchain/agents'
 
// 将工具的配置信息转为model接口里tools的结构化字段
// 由于对齐了接口规范,所以可以直接使用formatToOpenAITool函数
const modelWithTools = model.bind({
    tools: [...tools.map((tool: any) => formatToOpenAITool(tool))]
})
 
// 按顺序组合
const runnableAgent = RunnableSequence.from([
    // 包含了用户的指令,和将模型消息里的tool_calls format后得到的ToolMessage,和上下文聊天记录
    // 以上这些都会输入给prompt
    {
        [inputKey]: (i: { input: string; steps: AgentStep[] }) => i.input,
        agent_scratchpad: (i: { input: string; steps: ToolsAgentStep[] }) => formatToolAgentSteps(i.steps),
        [memoryKey]: async (_: { input: string; steps: AgentStep[] }) => {
            const messages = (await memory.getChatMessages(flowObj?.sessionId, true, chatHistory)) as BaseMessage[]
            return messages ?? []
        }
    },
    prompt,
    modelWithTools,
    new OpenAIToolsAgentOutputParser()
])
 
const executor = AgentExecutor.fromAgentAndTools({
    agent: runnableAgent,
    tools,
    returnIntermediateSteps: true,
    maxIterations: 5
})
 
executor.invoke({input: '明天是几月几号?'})
 
// tool_calls示例
{
    "tool_calls": [
      {
        "index": 0,
        "id": "GetDate:0",
        "type": "function",
        "function": {
          "name": "GetDate",
          "arguments": "{\n    \"task\": \"获取明天的日期\"\n}"
        }
      }
    ]
}