使用Langchain实现AI搜索，提升Langchain使用技巧，改造AI搜索实现----4

search-api 国内不能使用

使用zhipu 的 web-search-pro 进行搜索

import os
from typing import List
from pydantic import BaseModel
from langchain.prompts import PromptTemplate
from langchain_community.llms.tongyi import Tongyi
from langchain_core.runnables import RunnablePassthrough
from serpapi import client
# 搜索 使用智普的
import requests
from zhipuai import ZhipuAI

# 通义获取api key
import os
from dotenv import find_dotenv, load_dotenv

load_dotenv(find_dotenv())
DASHSCOPE_API_KEY = os.environ["DASHSCOPE_API_KEY"]

api_key = "fde24905ae3c5af19145593f767cdfde.NlFmUfzSqkJmrhb"


# # 调用web-search-pro获取搜索结果
def get_search_results(query: str):
    msg = [
        {
            "role": "user",
            "content": query,
        }
    ]
    tool = "web-search-pro"
    url = "https://open.bigmodel.cn/api/paas/v4/tools"
    data = {
        "tool": tool,
        "stream": False,
        "messages": msg
    }

    resp = requests.post(
        url,
        json=data,
        headers={'Authorization': api_key},
        timeout=300
    )
    res = eval(resp.content.decode())  # 转换成字典
    list_v = res.get("choices")
    for mes in list_v:
        value = mes["message"]
        tool_c = value.get("tool_calls")
    for tool_data in tool_c:
        result_v = tool_data.get("search_result")
    for dict_v in result_v:
        # print(dict_v["title"])
        # print(dict_v["content"])
        return dict_v


# get_search_results(query="易建联什么时候退役的")


# 1.定义模型
model = Tongyi(model_name='qwen-max', model_kwargs={'temperature': 0.01})

# 2 定义提示
prompt_tlp = PromptTemplate.from_template(
    "根据搜索引擎的结果，回答用户的问题。\nSEARCH_RESULTS:{search_result}\nUSER_QUERY:{query}"
)

# 创建搜索Chain
search_chain = {
                   "search_result": get_search_results,  # 引入搜索大模型结果计算
                   "query": RunnablePassthrough()  # 获取变量值。 相当于get_search_results, 获取的结果的变量
               } | prompt_tlp | model

# query = '2025年1月6日，福彩3D开奖号是多少？'
query = '清明上河图密码是什么意思？'

res = search_chain.invoke(query)

print(res)

问一个目前热播的事情

 

解耦搜索引擎

就是，加一个 pydantic 返回参数校验

定义输出，数据格式

class SearchResultItem(BaseModel):
    title: str
    link: str
    snippet: str  # 摘要

class SearchResults(BaseModel):
    results: List[SearchResultItem]

 代码

固定输出 title， link， content

 for dict_v in result_v:
        # print(dict_v["title"])
        # print(dict_v["content"])
        # return dict_v
        # --加一个使用pydantic校验，固定输出---
        search_results = SearchResults(
            results=[
                SearchResultItem(
                    title=dict_v['title'],
                    link=dict_v['link'],
                    content=dict_v['content']
                )
            ],
        )
        print(search_results)

 

序列化之后输出的数据，然后进行计算

from typing import List
from pydantic import BaseModel
from langchain.prompts import PromptTemplate
from langchain_community.llms.tongyi import Tongyi
from langchain_core.runnables import RunnablePassthrough
# 搜索 使用智普的
import requests
from zhipuai import ZhipuAI

# 通义获取api key
import os
from dotenv import find_dotenv, load_dotenv

load_dotenv(find_dotenv())
DASHSCOPE_API_KEY = os.environ["DASHSCOPE_API_KEY"]

api_key = "fde24905ae3c5af1914559f767cdfde.NlFmUfz3SqkJmrhb"


# pydantic，搜索结果数据模型model
class SearchResultItem(BaseModel):  # 存储单条搜索结果
    title: str
    link: str
    content: str  # 摘要

class SearchResults(BaseModel):       # 存储多条搜索结果
    results: List[SearchResultItem]


# # 调用web-search-pro获取搜索结果
def get_search_results(query: str) -> SearchResults:
    msg = [
        {
            "role": "user",
            "content": query,
        }
    ]
    tool = "web-search-pro"
    url = "https://open.bigmodel.cn/api/paas/v4/tools"
    data = {
        "tool": tool,
        "stream": False,
        "messages": msg
    }

    resp = requests.post(
        url,
        json=data,
        headers={'Authorization': api_key},
        timeout=300
    )
    res = eval(resp.content.decode())  # 转换成字典
    list_v = res.get("choices")
    for mes in list_v:
        value = mes["message"]
        tool_c = value.get("tool_calls")
    for tool_data in tool_c:
        result_v = tool_data.get("search_result")
    for dict_v in result_v:
        # print(dict_v["title"])
        # print(dict_v["content"])
        # return dict_v
        # -----------加一个使用pydantic校验，序列化，固定输出---
        search_results = SearchResults(
            results=[
                SearchResultItem(
                    title=dict_v['title'],
                    link=dict_v['link'],
                    content=dict_v['content']
                )
            ],
        )
        # print(search_results)
        return search_results



# get_search_results(query="梁朝伟多大了？")


# 1.定义模型
model = Tongyi(model_name='qwen-max', model_kwargs={'temperature': 0.01})

# 2 定义提示
prompt_tlp = PromptTemplate.from_template(
    """
    根据搜索引擎的结果，回答用户的问题。\n
    SEARCH_RESULTS: {search_result}\n
    USER_QUERY: {query}
    """
)

# 创建搜索Chain
search_chain = {
                   "search_result": lambda x: get_search_results(x),
                   "query": lambda x: x,
               } | prompt_tlp | model

# query = '2025年1月6日，福彩3D开奖号是多少？'
query = '清明上河图密码是什么意思？'

res = search_chain.invoke(query)

print(res)

end...