Merge remote-tracking branch 'origin/geo' into geo

src/agent/tool_divisions.py

+
+divisions = [
+    {
+    "province": "青海省",
+    "cities": [
+        {
+        "name": "西宁市",
+        "counties": ["城东区", "城中区", "城西区", "城北区", "大通回族土族自治县", "湟中区", "湟源县"]
+        },
+        {
+        "name": "海东市",
+        "counties": ["乐都区", "平安区", "民和回族土族自治县", "互助土族自治县", "化隆回族自治县", "循化撒拉族自治县"]
+        },
+        {
+        "name": "海北藏族自治州",
+        "counties": ["门源回族自治县", "祁连县", "海晏县", "刚察县"]
+        },
+        {
+        "name": "黄南藏族自治州",
+        "counties": ["同仁市", "尖扎县", "泽库县", "河南蒙古族自治县"]
+        },
+        {
+        "name": "海南藏族自治州",
+        "counties": ["共和县", "同德县", "贵德县", "兴海县", "贵南县"]
+        },
+        {
+        "name": "果洛藏族自治州",
+        "counties": ["玛沁县", "班玛县", "甘德县", "达日县", "久治县", "玛多县"]
+        },
+        {
+        "name": "玉树藏族自治州",
+        "counties": ["玉树市", "杂多县", "称多县", "治多县", "囊谦县", "曲麻莱县"]
+        },
+        {
+        "name": "海西蒙古族藏族自治州",
+        "counties": ["格尔木市", "德令哈市", "茫崖市", "乌兰县", "都兰县", "天峻县"]
+        }
+      ]
+    },
+    {
+    "province": "甘肃省",
+    "cities": [
+        {
+        "name": "兰州市",
+        "counties": ["城关区", "七里河区", "西固区", "安宁区", "红古区", "永登县", "皋兰县", "榆中县"]
+        },
+        {
+        "name": "嘉峪关市",
+        "counties": ["嘉峪关市"]
+        },
+        {
+        "name": "金昌市",
+        "counties": ["金川区", "永昌县"]
+        },
+        {
+        "name": "白银市",
+        "counties": ["白银区", "平川区", "靖远县", "会宁县", "景泰县"]
+        },
+        {
+        "name": "天水市",
+        "counties": ["秦州区", "麦积区", "清水县", "秦安县", "甘谷县", "武山县", "张家川回族自治县"]
+        },
+        {
+        "name": "武威市",
+        "counties": ["凉州区", "民勤县", "古浪县", "天祝藏族自治县"]
+        },
+        {
+        "name": "张掖市",
+        "counties": ["甘州区", "肃南裕固族自治县", "民乐县", "临泽县", "高台县", "山丹县"]
+        },
+        {
+        "name": "平凉市",
+        "counties": ["崆峒区", "泾川县", "灵台县", "崇信县", "华亭市", "庄浪县", "静宁县"]
+        },
+        {
+        "name": "酒泉市",
+        "counties": ["肃州区", "金塔县", "瓜州县", "肃北蒙古族自治县", "阿克塞哈萨克族自治县", "玉门市", "敦煌市"]
+        },
+        {
+        "name": "庆阳市",
+        "counties": ["西峰区", "庆城县", "环县", "华池县", "合水县", "正宁县", "宁县", "镇原县"]
+        },
+        {
+        "name": "定西市",
+        "counties": ["安定区", "通渭县", "陇西县", "渭源县", "临洮县", "漳县", "岷县"]
+        },
+        {
+        "name": "陇南市",
+        "counties": ["武都区", "成县", "文县", "宕昌县", "康县", "西和县", "礼县", "徽县", "两当县"]
+        },
+        {
+        "name": "临夏回族自治州",
+        "counties": ["临夏市", "临夏县", "康乐县", "永靖县", "广河县", "和政县", "东乡族自治县", "积石山保安族东乡族撒拉族自治县"]
+        },
+        {
+        "name": "甘南藏族自治州",
+        "counties": ["合作市", "临潭县", "卓尼县", "舟曲县", "迭部县", "玛曲县", "碌曲县", "夏河县"]
+        }
+      ]
+    }
+]
+
+
+from typing import Type
+from pydantic import BaseModel, Field
+from langchain_core.tools import BaseTool
+
+# 根据输入补全行政区划信息
+def complete_administrative_division(input_text, data):
+    for province in data:
+        if input_text in province['province']:
+            return {'province': province['province'], 'cities': [city['name'] for city in province['cities']]}
+        for city in province['cities']:
+            if input_text in city['name']:
+                return {
+                    '省': province['province'],
+                    '市': city['name'],
+                    '县(区)': city['counties']
+                }
+            for county in city['counties']:
+                if input_text in county:
+                    return {
+                        '省': province['province'],
+                        '市': city['name'],
+                        '县(区)': county
+                    }
+    return None
+
+
+class AdministrativeDivisionArgs(BaseModel):
+    input_text: str = Field(..., description="输入的行政区划信息,可以是省、市、县(区)三级中的最低一级的行政区划名称,例如用户输入'广东省广州市'，则此参数只需要取'广州市',用户输入'武汉市洪山区'，则次参数为'洪山区'")
+
+class AdministrativeDivision(BaseTool):
+    name = "administrative_division"
+    description = "根据输入补全行政区划信息，明确具体的省、市、县信息。比如输入县，补全所属省市，输入市则补全省级以及下辖所有县区"
+    args_schema: Type[BaseModel] = AdministrativeDivisionArgs
+
+    def _run(self, input_text: str) -> str:
+        result = complete_administrative_division(input_text, divisions)
+        return result
+    
+
+class CountryInfoArgs(BaseModel):
+    country: str = Field(...,description="县(区)这个级别的行政区划名称")
+
+# fake tool for testing
+class CountryInfo(BaseTool):
+    name = "country_info"
+    description = "查询特定县(区)相关的地质情况"
+    args_schema: Type[BaseModel] = CountryInfoArgs
+
+    def _run(self, country: str) -> str:
+        return f"{country} 的地质情况是：xxxxxx"
\ No newline at end of file

src/config/prompts.py

@@ -45,7 +45,7 @@ PROMPT_AGENT_HUMAN = """{input}\n\n{agent_scratchpad}\n (请注意，无论如
 ##################################################################################################################
 
 # 结合历史对话信息，对用户提问进行扩展，生成不同角度的多个提问。用于 RAG 场景中
-PROMPT_QUERY_EXTEND = """作为一个向量检索助手，你的任务是结合历史记录，从不同角度，为“原问题”生成三个不同版本的“检索词”，从而提高向量检索的语义丰富度，提高向量检索的精度。生成的问题要求指向对象清晰明确，并与“原问题语言相同”。例如：
+PROMPT_QUERY_EXTEND = """作为一个向量检索助手，你的任务是结合历史记录，从不同角度，为“原问题”生成三个不同版本的“检索词”，从而提高向量检索的语义丰富度，提高向量检索的精度。生成的问题要求指向对象清晰明确，并与“原问题语言相同”。如果问题中包含多个问询对象，可以将其拆分为多个问题。例如：
 历史记录: 
 '''
 '''
@@ -67,7 +67,7 @@ A: 当前对话是关于 Nginx 的介绍和使用等。
 Q: 报错 "no connection"
 A: 报错"no connection"可能是因为……
 '''
-原问题: 怎么解决
+原问题: 这是什么原因要怎么解决
 检索词: ["Nginx报错"no connection"如何解决？","造成'no connection'报错的原因。","Nginx提示'no connection'，要怎么办？"]
 ----------------
 历史记录: 
@@ -96,11 +96,9 @@ A: 关于 FatGPT 的介绍和使用等问题。
 ----------------
 历史记录:
 '''
-Q: FastGPT 如何收费？
-A: FastGPT 收费可以参考……
 '''
-原问题: 你知道 laf 么？
-检索词: ["laf 的官网地址是多少？","laf 的使用教程。","laf 有什么特点和优势。"]
+原问题: 北京和上海那个天气好？
+检索词: ["北京的天气情况如何","上海的天气情况如何","北京和上海的天气那个更舒适"]
 ----------------
 历史记录:
 '''

src/server/agent.py

+from typing import Any, List
+
+from langchain_core.prompts import PromptTemplate
+from langchain.agents import AgentExecutor, create_tool_calling_agent,create_structured_chat_agent
+from langchain.tools import BaseTool
+from langgraph.prebuilt import create_react_agent
+
+class Agent:
+    def __init__(self, llm, tools: List[BaseTool],prompt: PromptTemplate = None, verbose: bool = False):
+        self.llm = llm
+        self.tools = tools
+        self.prompt = prompt
+
+        if not prompt:
+            agent = create_react_agent(llm, tools,debug=verbose)
+            self.agent_executor = agent
+        else:
+            agent = create_structured_chat_agent(llm, tools, prompt)
+            self.agent_executor = AgentExecutor(agent=agent,tools=tools,verbose=verbose)        
+
+    def exec(self, prompt_args: dict = {}, stream: bool = False):
+        # if stream:
+        #     for step in self.agent_executor.stream(prompt_args):
+        #         yield step  
+        return self.agent_executor.invoke(input=prompt_args)
+        
+    def stream(self, prompt_args: dict = {}):
+        for step in self.agent_executor.stream(prompt_args):
+            yield step
\ No newline at end of file

test/agent_test.py

+import sys,os
+sys.path.append("../")
+
+from typing import List, Union, Type, Optional
+
+from langchain import hub
+import langchain_core
+from langchain_core.tools import tool, BaseTool
+from langchain_core.prompts import ChatPromptTemplate,PromptTemplate
+from langchain_core.prompts.chat import ChatPromptTemplate,HumanMessagePromptTemplate,SystemMessagePromptTemplate,MessagesPlaceholder
+from langchain_core.callbacks import CallbackManagerForToolRun
+from langchain_openai import ChatOpenAI
+from langchain_openai.embeddings import OpenAIEmbeddings
+from langchain.agents import AgentExecutor, create_tool_calling_agent,create_structured_chat_agent
+
+from pydantic import BaseModel, Field
+
+from src.server.agent import Agent
+from src.config.prompts import PROMPT_AGENT_SYS, PROMPT_AGENT_HUMAN
+
+from src.agent.tool_divisions import AdministrativeDivision, CountryInfo
+
+class CalcInput(BaseModel):
+    a: int = Field(...,description="第一个数")
+    b: int = Field(...,description="第二个数")
+
+class Calc(BaseTool):
+    name = "calc"
+    description = "一个简单的计算工具，可以计算两个数的和"
+    args_schema: Type[BaseModel] = CalcInput
+
+    def _run(
+        self, a: int, b: int, run_manager: Optional[CallbackManagerForToolRun] = None
+    ) -> str:
+        """Use the tool."""
+        print(f"Calculating {a} + {b}")
+        return a + b
+
+
+
+
+tools = [AdministrativeDivision()]
+
+llm = ChatOpenAI(
+    openai_api_key='xxxxxxxxxxxxx',
+    openai_api_base='http://192.168.10.14:8000/v1',
+    # openai_api_base='https://127.0.0.1:8000/v1',
+    model_name='Qwen2-7B',
+    verbose=True,
+    temperature=0,
+)
+
+# prompt = hub.pull("hwchase17/openai-functions-agent")
+input_variables=['agent_scratchpad', 'input', 'tool_names', 'tools'] 
+input_types={'chat_history': List[Union[langchain_core.messages.ai.AIMessage, langchain_core.messages.human.HumanMessage, langchain_core.messages.chat.ChatMessage, langchain_core.messages.system.SystemMessage, langchain_core.messages.function.FunctionMessage, langchain_core.messages.tool.ToolMessage]]} 
+messages=[
+  SystemMessagePromptTemplate(prompt=PromptTemplate(input_variables=['tool_names', 'tools'], template=PROMPT_AGENT_SYS)), 
+  MessagesPlaceholder(variable_name='chat_history', optional=True), 
+  HumanMessagePromptTemplate(prompt=PromptTemplate(input_variables=['agent_scratchpad', 'input'], template=PROMPT_AGENT_HUMAN))
+]
+
+prompt = ChatPromptTemplate(
+    input_variables=input_variables,
+    input_types=input_types,
+    messages=messages
+)
+
+def test_add():
+    tools = [Calc()]
+    agent = Agent(llm=llm, tools=tools, prompt=prompt, verbose=True)
+
+    res = agent.exec(prompt_args={"input": "what is 1 + 1?"})
+
+    # agent = create_structured_chat_agent(llm, tools, prompt)
+    # agent_executor = AgentExecutor(agent=agent,tools=tools,verbose=True,handle_parsing_errors=True)        
+    # res = agent_executor.invoke(input={"input": "what is 1 + 1?"})
+
+    # print(res)
+
+
+    # for step in agent.stream(prompt_args={"input": "what is 1 + 1?"}):
+    #     print("== step ==")
+    #     print(step)
+     
+def test_agent_division():
+    tools = [AdministrativeDivision(),CountryInfo()]
+    agent = Agent(llm=llm, tools=tools, prompt=prompt, verbose=True)
+    res = agent.exec(prompt_args={"input": "我想知道陇南市西和县和文县的降雨量谁的多"})
+    print(res)
+
+if __name__ == "__main__":
+    test_agent_division()
\ No newline at end of file

test/rag_test.py

@@ -29,7 +29,7 @@ def test_qaext():
 def test_chatextend():
     ext = ChatExtend(base_llm)
     message = [
-        ("我们明天去爬山吧", "好呀"),
+        ("明天去爬山怎么样", "好主意"),
         ("天气怎么样", "天气晴朗"),
     ]