{"id":1360,"date":"2024-05-06T09:33:48","date_gmt":"2024-05-06T09:33:48","guid":{"rendered":"https:\/\/www.nicekj.com\/?p=1360"},"modified":"2024-05-06T09:34:05","modified_gmt":"2024-05-06T09:34:05","slug":"langchainxiliejiaochengshiyonghuihuadailifuyullmchaonengli","status":"publish","type":"post","link":"https:\/\/www.nicekj.com\/langchainxiliejiaochengshiyonghuihuadailifuyullmchaonengli.html","title":{"rendered":"LangChain\u7cfb\u5217\u6559\u7a0b\uff1a\u4f7f\u7528\u4f1a\u8bdd\u4ee3\u7406\u8d4b\u4e88LLM\u8d85\u80fd\u529b"},"content":{"rendered":"

\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u975e\u5e38\u5f3a\u5927\uff0c\u4f46\u5b83\u4eec\u7f3a\u4e4f\u4e00\u4e9b\u201c\u6700\u7b28\u201d\u7684\u8ba1\u7b97\u673a\u7a0b\u5e8f\u53ef\u4ee5\u8f7b\u677e\u5904\u7406\u7684\u7279\u5b9a\u80fd\u529b\u3002\u903b\u8f91\u3001\u8ba1\u7b97\u548c\u641c\u7d22\u662f\u8ba1\u7b97\u673a\u901a\u5e38\u64c5\u957f\u7684\u9886\u57df\uff0c\u4f46LLMs\u5219\u8868\u73b0\u4e0d\u4f73\u3002<\/p>\n

\u8ba1\u7b97\u673a\u53ef\u4ee5\u89e3\u51b3\u975e\u5e38\u590d\u6742\u7684\u6570\u5b66\u95ee\u9898\uff0c\u7136\u800c\uff0c\u5982\u679c\u6211\u4eec\u8981\u6c42GPT-4\u544a\u8bc9\u6211\u4eec4.1 * 7.9\u7684\u7b54\u6848\uff0c\u5b83\u4f1a\u5931\u8d25\uff1a<\/p>\n

\n

\"image.png\"<\/figure>\n<\/p>\n

\u6839\u636e\u4e00\u4e2a\u7b80\u5355\u7684\u8ba1\u7b97\u5668\uff0c\u7b54\u6848\u662f19.357\uff0c\u56db\u820d\u4e94\u5165\u5230\u4e09\u4f4d\u5c0f\u6570\u3002\u4ee4\u4eba\u7740\u8ff7\u7684\u662f\uff0c\u4e00\u4e2a\u7b80\u5355\u7684\u8ba1\u7b97\u5668\u7a0b\u5e8f\u53ef\u4ee5\u505a\u5230\u8fd9\u4e00\u70b9\uff0c\u4f46\u4e00\u4e2a\u6781\u5176\u590d\u6742\u7684AI\u5f15\u64ce\u5931\u8d25\u4e86\u3002<\/p>\n

\u8fd9\u8fd8\u4e0d\u662f\u5168\u90e8\u3002\u5982\u679c\u6211\u95eeGPT-4\uff0c\u201cHow do I use the LLMChain in LangChain?\u201d\u5b83\u518d\u6b21\u9047\u5230\u56f0\u96be\uff1a<\/p>\n

\n

\"image.png\"<\/figure>\n<\/p>\n

LangChain\u786e\u5b9e\u662f\u4e00\u4e2a\u533a\u5757\u94fe\u9879\u76ee[1] [2]\u3002\u7136\u800c\uff0c\u4f3c\u4e4e\u6ca1\u6709\u4efb\u4f55\u201cLLMChain\u201d\u7ec4\u4ef6\u6216\u201cLANG\u4ee3\u5e01\u201d\u2014\u2014\u8fd9\u4e24\u8005\u90fd\u662f\u5e7b\u89c9\u3002<\/p>\n

GPT-4\u65e0\u6cd5\u544a\u8bc9\u6211\u4eec\u6709\u5173LangChain\u7684\u539f\u56e0\u662f\u5b83\u4e0e\u5916\u90e8\u4e16\u754c\u6ca1\u6709\u8054\u7cfb\u3002\u5b83\u552f\u4e00\u7684\u77e5\u8bc6\u6765\u81ea\u4e8e\u5b83\u4ece\u8bad\u7ec3\u6570\u636e\u4e2d\u6355\u83b7\u7684\u5185\u5bb9\uff0c\u800c\u8fd9\u4e9b\u6570\u636e\u622a\u6b62\u52302021\u5e74\u672b\u3002<\/p>\n

\u7531\u4e8e\u4eca\u5929\u7684LLMs\u5b58\u5728\u91cd\u5927\u5f31\u70b9\uff0c\u6211\u4eec\u5fc5\u987b\u627e\u5230\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\u7684\u65b9\u6cd5\u3002\u4e00\u7ec4\u6f5c\u5728\u7684\u89e3\u51b3\u65b9\u6848\u4ee5\u201c\u4ee3\u7406\u4eba\u201d\u7684\u5f62\u5f0f\u51fa\u73b0\u3002<\/p>\n

\u8fd9\u4e9b\u4ee3\u7406\u4eba\u4e0d\u4ec5\u89e3\u51b3\u4e86\u6211\u4eec\u4e0a\u9762\u770b\u5230\u7684\u95ee\u9898\uff0c\u8fd8\u89e3\u51b3\u4e86\u8bb8\u591a\u5176\u4ed6\u95ee\u9898\u3002\u4e8b\u5b9e\u4e0a\uff0c\u6dfb\u52a0\u4ee3\u7406\u4eba\u5728\u63d0\u5347LLM\u80fd\u529b\u65b9\u9762\u5177\u6709\u51e0\u4e4e\u65e0\u9650\u7684\u6f5c\u529b\u3002<\/p>\n

\u5728\u672c\u7ae0\u4e2d\uff0c\u6211\u4eec\u5c06\u8ba8\u8bba\u4ee3\u7406\u4eba\u3002\u6211\u4eec\u5c06\u4e86\u89e3\u5b83\u4eec\u662f\u4ec0\u4e48\uff0c\u5b83\u4eec\u5982\u4f55\u5de5\u4f5c\uff0c\u4ee5\u53ca\u5982\u4f55\u5728LangChain\u5e93\u4e2d\u4f7f\u7528\u5b83\u4eec\u6765\u8d4b\u4e88\u6211\u4eec\u7684LLMs\u8d85\u80fd\u529b\u3002<\/p>\n

\u4ec0\u4e48\u662f\u4ee3\u7406\uff1f<\/h1>\n

\u6211\u4eec\u53ef\u4ee5\u5c06\u4ee3\u7406\u4eba\u770b\u4f5c\u662fLLMs\u7684\u201c\u5de5\u5177\u201d\u3002\u5c31\u50cf\u4eba\u7c7b\u4f1a\u4f7f\u7528\u8ba1\u7b97\u5668\u8fdb\u884c\u6570\u5b66\u8fd0\u7b97\u6216\u8fdb\u884c\u8c37\u6b4c\u641c\u7d22\u83b7\u53d6\u4fe1\u606f\u4e00\u6837\uff0c\u4ee3\u7406\u4eba\u5141\u8bb8LLMs\u6267\u884c\u7c7b\u4f3c\u7684\u64cd\u4f5c\u3002<\/p>\n

\n

\"image.png\"<\/figure>\n<\/p>\n

\u4f7f\u7528\u4ee3\u7406\uff0cLLM\u53ef\u4ee5\u7f16\u5199\u548c\u6267\u884cPython\u4ee3\u7801\u3002\u5b83\u53ef\u4ee5\u641c\u7d22\u4fe1\u606f\uff0c\u751a\u81f3\u67e5\u8be2SQL\u6570\u636e\u5e93\u3002<\/p>\n

\u8ba9\u6211\u4eec\u770b\u4e00\u4e2a\u7b80\u5355\u7684\u793a\u4f8b\u3002\u6211\u4eec\u5c06\u4ece\u4e00\u4e2a\u201c\u96f6-shot\u201d\u4ee3\u7406\u5f00\u59cb\uff08\u7a0d\u540e\u8be6\u7ec6\u4ecb\u7ecd\uff09\uff0c\u5141\u8bb8\u6211\u4eec\u7684LLM\u4f7f\u7528\u8ba1\u7b97\u5668\u3002<\/p>\n

\u4ee3\u7406\u548c\u5de5\u5177<\/h2>\n

\u8981\u4f7f\u7528\u4ee3\u7406\uff0c\u6211\u4eec\u9700\u8981\u4e09\u6837\u4e1c\u897f\uff1a<\/p>\n

    \n
  1. \u57fa\u7840LLM\uff0c<\/li>\n
  2. \u6211\u4eec\u5c06\u8981\u8fdb\u884c\u4ea4\u4e92\u7684\u5de5\u5177\uff0c<\/li>\n
  3. \u7528\u4e8e\u63a7\u5236\u4ea4\u4e92\u7684\u4ee3\u7406\u3002<\/li>\n<\/ol>\n

    \u8ba9\u6211\u4eec\u5f00\u59cb\u5b89\u88c5langchain\u5e76\u521d\u59cb\u5316\u6211\u4eec\u7684\u57fa\u7840LLM\u3002<\/p>\n

    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain import OpenAI<\/span>\n<\/span>\nllm<\/span> = OpenAI(<\/span>\n    openai_api_key<\/span>=\"OPENAI_API_KEY\"<\/span>,<\/span>\n    temperature<\/span>=0<\/span>,<\/span>\n    model_name<\/span>=\"text-davinci-003\"<\/span><\/span>\n)<\/span>\n<\/code><\/pre>\n
    \u73b0\u5728\u521d\u59cb\u5316\u8ba1\u7b97\u5668\u5de5\u5177\u3002\u5728\u521d\u59cb\u5316\u5de5\u5177\u65f6\uff0c\u6211\u4eec\u53ef\u4ee5\u521b\u5efa\u81ea\u5b9a\u4e49\u5de5\u5177\u6216\u52a0\u8f7d\u9884\u6784\u5efa\u5de5\u5177\u3002\u65e0\u8bba\u54ea\u79cd\u60c5\u51b5\uff0c\u201c\u5de5\u5177\u201d\u90fd\u662f\u4e00\u4e2a\u5177\u6709\u5de5\u5177\u540d\u79f0\u548c\u63cf\u8ff0\u7684\u5b9e\u7528\u7a0b\u5e8f\u94fe\u3002<\/p>\n
    \u4f8b\u5982\uff0c\u6211\u4eec\u53ef\u4ee5\u4ece\u73b0\u6709\u7684llm_math\u94fe\u521b\u5efa\u4e00\u4e2a\u65b0\u7684\u8ba1\u7b97\u5668\u5de5\u5177\uff1a<\/p>\n
    In[3]:<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain.chains import LLMMathChain<\/span>\nfrom langchain.agents import Tool<\/span>\n<\/span>\nllm_math<\/span> = LLMMathChain(llm=llm)<\/span>\n<\/span>\n# initialize the math tool<\/span><\/span>\nmath_tool<\/span> = Tool(<\/span>\n    name<\/span>='Calculator'<\/span>,<\/span>\n    func<\/span>=llm_math.run,<\/span>\n    description<\/span>='Useful for when you need to answer questions about math.'<\/span><\/span>\n)<\/span>\n# when giving tools to LLM, we must pass as list of tools<\/span><\/span>\ntools<\/span> = [math_tool]<\/span>\n<\/code><\/pre>\n
    In[4]:<\/p>\n
    <\/div>
    css<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>tools[0]<\/span>.name<\/span>, tools[0]<\/span>.description<\/span><\/span>\n<\/code><\/pre>\n
    Out[4]:<\/p>\n
    <\/div>
    rust<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>('Calculator<\/span>', 'Useful<\/span> for<\/span> when<\/span> you need to answer questions about math.')<\/span>\n<\/code><\/pre>\n
    \u5728\u4f7f\u7528\u81ea\u5b9a\u4e49\u5de5\u5177\u65f6\uff0c\u6211\u4eec\u5fc5\u987b\u9075\u5faa\u8fd9\u4e2a\u8fc7\u7a0b\u3002\u7136\u800c\uff0c\u9884\u6784\u5efa\u7684llm_math\u5de5\u5177\u6267\u884c\u76f8\u540c\u7684\u64cd\u4f5c\u3002\u56e0\u6b64\uff0c\u6211\u4eec\u53ef\u4ee5\u50cf\u4e0a\u9762\u90a3\u6837\u505a\u76f8\u540c\u7684\u64cd\u4f5c\uff0c\u5982\u4e0b\u6240\u793a\uff1a<\/p>\n
    In[5]:<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain.agents import load_tools<\/span>\n<\/span>\ntools<\/span> = load_tools(<\/span>\n    ['llm-math']<\/span>,<\/span>\n    llm<\/span>=llm<\/span>\n)<\/span>\n<\/code><\/pre>\n
    In[6]:<\/p>\n
    <\/div>
    css<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>tools[0]<\/span>.name<\/span>, tools[0]<\/span>.description<\/span><\/span>\n<\/code><\/pre>\n
    Out[6]:<\/p>\n
    <\/div>
    rust<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>('Calculator<\/span>', 'Useful<\/span> for<\/span> when<\/span> you need to answer questions about math.')<\/span>\n<\/code><\/pre>\n
    \u5f53\u7136\uff0c\u53ea\u6709\u5728\u6211\u4eec\u7684\u7528\u4f8b\u4e2d\u5b58\u5728\u9884\u6784\u5efa\u7684\u5de5\u5177\u65f6\uff0c\u6211\u4eec\u624d\u80fd\u91c7\u7528\u7b2c\u4e8c\u79cd\u65b9\u6cd5\u3002<\/p>\n
    \u73b0\u5728\uff0c\u6211\u4eec\u6709\u4e86LLM\u548c\u5de5\u5177\uff0c\u4f46\u8fd8\u6ca1\u6709\u4ee3\u7406\u3002\u8981\u521d\u59cb\u5316\u4e00\u4e2a\u7b80\u5355\u7684\u4ee3\u7406\uff0c\u6211\u4eec\u53ef\u4ee5\u6267\u884c\u4ee5\u4e0b\u6b65\u9aa4\uff1a<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain.agents import initialize_agent<\/span>\n<\/span>\nzero_shot_agent<\/span> = initialize_agent(<\/span>\n    agent<\/span>=\"zero-shot-react-description\"<\/span>,<\/span>\n    tools<\/span>=tools,<\/span>\n    llm<\/span>=llm,<\/span>\n    verbose<\/span>=True<\/span>,<\/span>\n    max_iterations<\/span>=3<\/span><\/span>\n)<\/span>\n<\/code><\/pre>\n
    \u8fd9\u91cc\u4f7f\u7528\u7684\u4ee3\u7406\u662f\u4e00\u4e2a\u201c\u96f6\u6b21\u53cd\u5e94\u63cf\u8ff0\u201d\u4ee3\u7406\u3002\u96f6\u6b21\u53cd\u5e94\u610f\u5473\u7740\u4ee3\u7406\u4ec5\u5728\u5f53\u524d\u64cd\u4f5c\u4e0a\u6267\u884c\u529f\u80fd – \u5b83\u6ca1\u6709\u8bb0\u5fc6\u3002\u5b83\u4f7f\u7528ReAct\u6846\u67b6\u6765\u51b3\u5b9a\u4f7f\u7528\u54ea\u4e2a\u5de5\u5177\uff0c\u4ec5\u57fa\u4e8e\u5de5\u5177\u7684\u63cf\u8ff0\u3002<\/p>\n
    \u5728\u672c\u7ae0\u4e2d\uff0c\u6211\u4eec\u4e0d\u4f1a\u8ba8\u8bbaReAct\u6846\u67b6\uff0c\u4f46\u4f60\u53ef\u4ee5\u5c06\u5176\u770b\u4f5c\u662fLLM\u53ef\u4ee5\u5728\u63a8\u7406\u548c\u884c\u52a8\u6b65\u9aa4\u4e4b\u95f4\u5faa\u73af\u7684\u65b9\u5f0f\uff0c\u4ee5\u4fbf\u8bc6\u522b\u7b54\u6848\u7684\u591a\u6b65\u8fc7\u7a0b\u3002<\/p>\n
    \u521d\u59cb\u5316\u4e86\u6211\u4eec\u7684\u4ee3\u7406\u540e\uff0c\u6211\u4eec\u53ef\u4ee5\u5f00\u59cb\u4f7f\u7528\u5b83\u3002\u8ba9\u6211\u4eec\u5c1d\u8bd5\u4e00\u4e9b\u63d0\u793a\uff0c\u770b\u770b\u4ee3\u7406\u5982\u4f55\u54cd\u5e94\u3002<\/p>\n
    In[8]:<\/p>\n
    <\/div>
    scss<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>zero_shot_agent<\/span>(\"what is (4.5<\/span>*2.1<\/span>)^2.2<\/span>?\")<\/span>\n<\/code><\/pre>\n
    Out[8]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> AgentExecutor chain...\u001b[0<\/span>m<\/span>\n\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I need to<\/span> calculate this expression<\/span>\nAction:<\/span> Calculator<\/span>\nAction Input: (4.5<\/span>*2.1<\/span>)^2.2<\/span>\u001b[0<\/span>m<\/span>\nObservation:<\/span> \u001b[36<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>mAnswer: 139.94261298333066<\/span><\/span>\n\u001b[0<\/span>m<\/span>\nThought:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I now know the final answer<\/span>\nFinal Answer: 139.94261298333066<\/span>\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m> Finished chain.\u001b[0<\/span>m<\/span>\n<\/code><\/pre>\n
    Out[8]:<\/p>\n
    <\/div>
    css<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>{'input<\/span>': 'what is (4.5*2.1)^2.2?'<\/span>, 'output'<\/span>: '139.94261298333066'<\/span>}<\/span>\n<\/code><\/pre>\n
    In[9]:<\/p>\n
    <\/div>
    scss<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>(4.5<\/span>*2.1<\/span>)**2.2<\/span><\/span>\n<\/code><\/pre>\n
    Out[9]:<\/p>\n
    <\/div>
    <\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>139.94261298333066<\/span>\n<\/code><\/pre>\n
    \u8fd9\u91cc\u7684\u7b54\u6848\u662f\u6b63\u786e\u7684\u3002\u8ba9\u6211\u4eec\u5c1d\u8bd5\u53e6\u4e00\u4e2a\uff1a<\/p>\n
    In[10]:<\/p>\n
    <\/div>
    erlang<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>zero_shot_agent<\/span>(\"if Mary has four apples and Giorgio brings two and a half apple \"<\/span><\/span><\/span><\/span>\n                \"boxes (apple box contains eight apples), how many apples do we \"<\/span><\/span>\n                \"have?\"<\/span>)<\/span>\n<\/span>\n<\/code><\/pre>\n
    Out[10]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> AgentExecutor chain...\u001b[0<\/span>m<\/span>\n\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I need to<\/span> figure out how many apples are in<\/span> the boxes<\/span>\nAction:<\/span> Calculator<\/span>\nAction Input: 8<\/span> * 2.5<\/span>\u001b[0<\/span>m<\/span>\nObservation:<\/span> \u001b[36<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>mAnswer: 20.0<\/span><\/span>\n\u001b[0<\/span>m<\/span>\nThought:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I need to<\/span> add the apples Mary has to<\/span> the apples in<\/span> the boxes<\/span>\nAction:<\/span> Calculator<\/span>\nAction Input: 4<\/span> + 20.0<\/span>\u001b[0<\/span>m<\/span>\nObservation:<\/span> \u001b[36<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>mAnswer: 24.0<\/span><\/span>\n\u001b[0<\/span>m<\/span>\nThought:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I now know the final answer<\/span>\nFinal Answer: We have 24<\/span> apples.\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m> Finished chain.\u001b[0<\/span>m<\/span>\n<\/code><\/pre>\n
    Out[10]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>{'input': 'if Mary has four apples and Giorgio brings two and a half apple boxes (apple box contains eight apples), how many apples do we have?',<\/span><\/span>\n 'output': 'We have 24 apples.'}<\/span><\/span>\n<\/code><\/pre>\n
    \u770b\u8d77\u6765\u4e0d\u9519\uff01\u4f46\u662f\uff0c\u5982\u679c\u6211\u4eec\u51b3\u5b9a\u95ee\u4e00\u4e2a\u975e\u6570\u5b66\u95ee\u9898\u5462\uff1f\u5982\u679c\u6211\u4eec\u95ee\u4e00\u4e2a\u7b80\u5355\u7684\u5e38\u8bc6\u95ee\u9898\u5462\uff1f<\/p>\n
    In[11]:<\/p>\n
    <\/div>
    scss<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>zero_shot_agent<\/span>(\"what is the capital of Norway?\")<\/span>\n<\/code><\/pre>\n
    Out[11]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> AgentExecutor chain...\u001b[0<\/span>m<\/span>\n\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I need to<\/span> look up the answer<\/span>\nAction:<\/span> Look up<\/span>\nAction Input: Capital of<\/span> Norway\u001b[0<\/span>m<\/span>\nObservation:<\/span> Look up is<\/span> not<\/span> a valid tool, try<\/span> another one.<\/span>\nThought:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I need to<\/span> find the answer using<\/span> a tool<\/span>\nAction:<\/span> Calculator<\/span>\nAction Input: N\/A\u001b[0<\/span>m<\/span>\n<\/code><\/pre>\n
    \u6211\u4eec\u9047\u5230\u4e86\u4e00\u4e2a\u9519\u8bef\u3002\u95ee\u9898\u5728\u4e8e\u4ee3\u7406\u4eba\u4e0d\u65ad\u5c1d\u8bd5\u4f7f\u7528\u5de5\u5177\u3002\u7136\u800c\uff0c\u6211\u4eec\u7684\u4ee3\u7406\u4eba\u53ea\u5305\u542b\u4e00\u4e2a\u5de5\u5177 – \u8ba1\u7b97\u5668\u3002<\/p>\n
    \u5e78\u8fd0\u7684\u662f\uff0c\u6211\u4eec\u53ef\u4ee5\u901a\u8fc7\u4e3a\u4ee3\u7406\u4eba\u6dfb\u52a0\u66f4\u591a\u5de5\u5177\u6765\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff01\u8ba9\u6211\u4eec\u6dfb\u52a0\u4e00\u4e2a\u7b80\u5355\u660e\u4e86\u7684LLM\u5de5\u5177\uff1a<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain.prompts import PromptTemplate<\/span>\nfrom langchain.chains import LLMChain<\/span>\n<\/span>\nprompt<\/span> = PromptTemplate(<\/span>\n    input_variables<\/span>=[\"query\"<\/span>],<\/span>\n    template<\/span>=\"{query}\"<\/span><\/span>\n)<\/span>\n<\/span>\nllm_chain<\/span> = LLMChain(llm=llm, prompt=prompt)<\/span>\n<\/span>\n# initialize the LLM tool<\/span><\/span>\nllm_tool<\/span> = Tool(<\/span>\n    name<\/span>='Language Model'<\/span>,<\/span>\n    func<\/span>=llm_chain.run,<\/span>\n    description<\/span>='use this tool for general purpose queries and logic'<\/span><\/span>\n)<\/span>\n<\/code><\/pre>\n
    \u6709\u4e86\u8fd9\u4e2a\uff0c\u6211\u4eec\u6709\u4e86\u4e00\u4e2a\u65b0\u7684\u901a\u7528LLM\u5de5\u5177\u3002\u6211\u4eec\u53ea\u9700\u5c06\u5b83\u6dfb\u52a0\u5230\u5de5\u5177\u5217\u8868\u4e2d\uff0c\u7136\u540e\u91cd\u65b0\u521d\u59cb\u5316\u4ee3\u7406\u4eba\uff1a<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>tools.append(llm_tool)<\/span>\n<\/span>\n# reinitialize the agent<\/span><\/span>\nzero_shot_agent<\/span> = initialize_agent(<\/span>\n    agent<\/span>=\"zero-shot-react-description\"<\/span>,<\/span>\n    tools<\/span>=tools,<\/span>\n    llm<\/span>=llm,<\/span>\n    verbose<\/span>=True<\/span>,<\/span>\n    max_iterations<\/span>=3<\/span><\/span>\n)<\/span>\n<\/code><\/pre>\n
    \u73b0\u5728\u6211\u4eec\u53ef\u4ee5\u5411\u4ee3\u7406\u4eba\u63d0\u95ee\u5173\u4e8e\u6570\u5b66\u548c\u5e38\u8bc6\u7684\u95ee\u9898\u3002\u8ba9\u6211\u4eec\u5c1d\u8bd5\u4ee5\u4e0b\u95ee\u9898\uff1a<\/p>\n
    In[15]:<\/p>\n
    <\/div>
    scss<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>zero_shot_agent<\/span>(\"what is the capital of Norway?\")<\/span>\n<\/code><\/pre>\n
    Out[15]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> AgentExecutor chain...\u001b[0<\/span>m<\/span>\n\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I need to<\/span> find out what the capital of<\/span> Norway is<\/span><\/span>\nAction:<\/span> Language Model<\/span>\nAction Input: What is<\/span> the capital of<\/span> Norway?\u001b[0<\/span>m<\/span>\nObservation:<\/span> \u001b[33<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m<\/span>\n<\/span>\nThe capital of<\/span> Norway is<\/span> Oslo.\u001b[0<\/span>m<\/span>\nThought:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I now know the final answer<\/span>\nFinal Answer: The capital of<\/span> Norway is<\/span> Oslo.\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m> Finished chain.\u001b[0<\/span>m<\/span>\n<\/code><\/pre>\n
    Out[15]:<\/p>\n
    <\/div>
    css<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>{'input<\/span>': 'what is the capital of Norway?'<\/span>,<\/span>\n 'output'<\/span>: 'The capital of Norway is Oslo.'<\/span>}<\/span>\n<\/code><\/pre>\n
    \u73b0\u5728\u6211\u4eec\u5f97\u5230\u4e86\u6b63\u786e\u7684\u7b54\u6848\uff01\u6211\u4eec\u53ef\u4ee5\u63d0\u51fa\u7b2c\u4e00\u4e2a\u95ee\u9898\uff1a<\/p>\n
    In[16]:<\/p>\n
    <\/div>
    scss<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>zero_shot_agent<\/span>(\"what is (4.5<\/span>*2.1<\/span>)^2.2<\/span>?\")<\/span>\n<\/code><\/pre>\n
    Out[16]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> AgentExecutor chain...\u001b[0<\/span>m<\/span>\n\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I need to<\/span> calculate this expression<\/span>\nAction:<\/span> Calculator<\/span>\nAction Input: (4.5<\/span>*2.1<\/span>)^2.2<\/span>\u001b[0<\/span>m<\/span>\nObservation:<\/span> \u001b[36<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>mAnswer: 139.94261298333066<\/span><\/span>\n\u001b[0<\/span>m<\/span>\nThought:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I now know the final answer<\/span>\nFinal Answer: 139.94261298333066<\/span>\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m> Finished chain.\u001b[0<\/span>m<\/span>\n<\/code><\/pre>\n
    Out[16]:<\/p>\n
    <\/div>
    css<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>{'input<\/span>': 'what is (4.5*2.1)^2.2?'<\/span>, 'output'<\/span>: '139.94261298333066'<\/span>}<\/span>\n<\/code><\/pre>\n
    \u800c\u4ee3\u7406\u660e\u767d\u5b83\u5fc5\u987b\u53c2\u8003\u8ba1\u7b97\u5668\u5de5\u5177\uff0c\u5b83\u4e5f\u505a\u5230\u4e86 – \u7ed9\u4e86\u6211\u4eec\u6b63\u786e\u7684\u7b54\u6848\u3002<\/p>\n
    \u5b8c\u6210\u8fd9\u4e00\u90e8\u5206\u540e\uff0c\u6211\u4eec\u5e94\u8be5\u7406\u89e3\u4e86\u4f7f\u7528\u4e0d\u540c\u5de5\u5177\u8bbe\u8ba1\u548c\u63d0\u793a\u4ee3\u7406\u7684\u5de5\u4f5c\u6d41\u7a0b\u3002\u73b0\u5728\u8ba9\u6211\u4eec\u7ee7\u7eed\u4e86\u89e3\u53ef\u7528\u4e8e\u6211\u4eec\u7684\u4e0d\u540c\u7c7b\u578b\u7684\u4ee3\u7406\u548c\u5de5\u5177\u3002<\/p>\n
    \u4ee3\u7406\u7c7b\u578b<\/h1>\n
    LangChain\u63d0\u4f9b\u4e86\u51e0\u79cd\u7c7b\u578b\u7684\u4ee3\u7406\u3002\u5728\u672c\u8282\u4e2d\uff0c\u6211\u4eec\u5c06\u68c0\u67e5\u4e00\u4e9b\u6700\u5e38\u89c1\u7684\u4ee3\u7406\u7c7b\u578b\u3002<\/p>\n
    \u96f6\u6b21\u53cd\u5e94 (Zero Shot ReAct)<\/h2>\n
    \u6211\u4eec\u5c06\u4ece\u4e4b\u524d\u770b\u5230\u7684\u4ee3\u7406\u4eba\u5f00\u59cb\uff0c\u5373\u96f6\u6b21\u53cd\u5e94\u63cf\u8ff0\uff08zero-shot-react-description\uff09\u4ee3\u7406\u4eba\u3002<\/p>\n
    \u5982\u524d\u6240\u8ff0\uff0c\u6211\u4eec\u4f7f\u7528\u8fd9\u4e2a\u4ee3\u7406\u4eba\u6765\u6267\u884c\u4e00\u4e9b\u201c\u96f6\u6b21\u53cd\u5e94\u201d\u4efb\u52a1\u3002\u8fd9\u610f\u5473\u7740\u4ee3\u7406\u4eba\u8003\u8651\u4e0e\u4ee3\u7406\u4eba\u7684\u4e00\u6b21\u5355\u4e00\u4ea4\u4e92 – \u5b83\u6ca1\u6709\u8bb0\u5fc6\u3002<\/p>\n
    \u8ba9\u6211\u4eec\u521b\u5efa\u4e00\u4e2a\u5de5\u5177\u5217\u8868\uff0c\u4ee5\u4f9b\u4ee3\u7406\u4eba\u4f7f\u7528\u3002\u6211\u4eec\u5c06\u5305\u62ec\u6211\u4eec\u5728\u8fd9\u91cc\u5b9a\u4e49\u7684llm-math\u5de5\u5177\u548cSQL\u6570\u636e\u5e93\u5de5\u5177\u3002<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>tools<\/span> = load_tools(<\/span>\n    [\"llm-math\"]<\/span>, <\/span>\n    llm<\/span>=llm<\/span>\n)<\/span>\n<\/span>\n# add our custom SQL db tool<\/span><\/span>\ntools.append(sql_tool)<\/span>\n<\/code><\/pre>\n
    \u6211\u4eec\u53ef\u4ee5\u8fd9\u6837\u521d\u59cb\u5316\u96f6\u6b21\u53cd\u5e94\u63cf\u8ff0\u4ee3\u7406\u4eba\uff1a<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain.agents import initialize_agent<\/span>\n<\/span>\nzero_shot_agent<\/span> = initialize_agent(<\/span>\n    agent<\/span>=\"zero-shot-react-description\"<\/span>, <\/span>\n    tools<\/span>=tools, <\/span>\n    llm<\/span>=llm,<\/span>\n    verbose<\/span>=True<\/span>,<\/span>\n    max_iterations<\/span>=3<\/span>,<\/span>\n)<\/span>\n<\/code><\/pre>\n
    \u4e3a\u4e86\u8ba9SQL\u6570\u636e\u5e93\u5de5\u5177\u6709\u4e00\u4e9b\u80cc\u666f\u4fe1\u606f\uff0c\u6211\u4eec\u5c06\u4f7f\u7528\u5b83\u6765\u67e5\u8be2\u4e00\u4e2a\u7c7b\u4f3c\u4e8e\u8fd9\u6837\u7684\u201c\u80a1\u7968\u6570\u636e\u5e93\u201d\uff1a<\/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
    obs_id<\/th>stock_ticker<\/th>price<\/th>data<\/th><\/tr><\/thead>
    1<\/td>\u2018ABC\u2019<\/td>200<\/td>1 Jan 23<\/td><\/tr>
    2<\/td>\u2018ABC\u2019<\/td>208<\/td>2 Jan 23<\/td><\/tr>
    3<\/td>\u2018ABC\u2019<\/td>232<\/td>3 Jan 23<\/td><\/tr>
    4<\/td>\u2018ABC\u2019<\/td>225<\/td>4 Jan 23<\/td><\/tr>
    5<\/td>\u2018ABC\u2019<\/td>226<\/td>5 Jan 23<\/td><\/tr>
    6<\/td>\u2018XYZ\u2019<\/td>810<\/td>1 Jan 23<\/td><\/tr>
    7<\/td>\u2018XYZ\u2019<\/td>803<\/td>2 Jan 23<\/td><\/tr>
    8<\/td>\u2018XYZ\u2019<\/td>798<\/td>3 Jan 23<\/td><\/tr>
    9<\/td>\u2018XYZ\u2019<\/td>795<\/td>4 Jan 23<\/td><\/tr>
    10<\/td>\u2018XYZ\u2019<\/td>791<\/td>5 Jan 23<\/td><\/tr><\/tbody><\/table>\n
    \u73b0\u5728\u6211\u4eec\u53ef\u4ee5\u5f00\u59cb\u95ee\u5173\u4e8e\u8fd9\u4e2aSQL\u6570\u636e\u5e93\u7684\u95ee\u9898\uff0c\u5e76\u901a\u8fc7\u8ba1\u7b97\u5668\u5de5\u5177\u8fdb\u884c\u8ba1\u7b97\u3002<\/p>\n
    In[16]:<\/p>\n
    <\/div>
    makefile<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>result = zero_shot_agent(<\/span>\n    \"What is the multiplication of the ratio between stock prices for 'ABC' \"<\/span><\/span>\n    \"and 'XYZ' in January 3rd and the ratio between the same stock prices in \"<\/span><\/span>\n    \"January the 4th?\"<\/span><\/span>\n)<\/span>\n<\/code><\/pre>\n
    Out[16]:<\/p>\n
    <\/div>
    sql<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\n\u001b[1<\/span>m><\/span> Entering new<\/span> AgentExecutor chain...\u001b[0<\/span>m<\/span>\n\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I need to<\/span> compare the stock prices of<\/span> 'ABC'<\/span> and<\/span> 'XYZ'<\/span> on<\/span> two different days<\/span>\nAction: Stock DB<\/span>\nAction Input: Stock prices of<\/span> 'ABC'<\/span> and<\/span> 'XYZ'<\/span> on<\/span> January 3<\/span>rd and<\/span> January 4<\/span>th\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m><\/span> Entering new<\/span> SQLDatabaseChain chain...\u001b[0<\/span>m<\/span>\nStock prices of<\/span> 'ABC'<\/span> and<\/span> 'XYZ'<\/span> on<\/span> January 3<\/span>rd and<\/span> January 4<\/span>th <\/span>\nSQLQuery:\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m SELECT<\/span> stock_ticker, price, date<\/span> FROM<\/span> stocks WHERE<\/span> (stock_ticker =<\/span> 'ABC'<\/span> OR<\/span> stock_ticker =<\/span> 'XYZ'<\/span>) AND<\/span> (date<\/span> =<\/span> '2023-01-03'<\/span> OR<\/span> date<\/span> =<\/span> '2023-01-04'<\/span>)\u001b[0<\/span>m<\/span>\nSQLResult: \u001b[33<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m[('ABC'<\/span>, 232.0<\/span>, '2023-01-03'<\/span>), ('ABC'<\/span>, 225.0<\/span>, '2023-01-04'<\/span>), ('XYZ'<\/span>, 798.0<\/span>, '2023-01-03'<\/span>), ('XYZ'<\/span>, 795.0<\/span>, '2023-01-04'<\/span>)]\u001b[0<\/span>m<\/span>\nAnswer:\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m The stock prices of<\/span> 'ABC'<\/span> and<\/span> 'XYZ'<\/span> on<\/span> January 3<\/span>rd and<\/span> January 4<\/span>th were 232.0<\/span> and<\/span> 798.0<\/span> respectively for<\/span> 'ABC'<\/span> and<\/span> 'XYZ'<\/span> on<\/span> January 3<\/span>rd, and<\/span> 225.0<\/span> and<\/span> 795.0<\/span> respectively for<\/span> 'ABC'<\/span> and<\/span> 'XYZ'<\/span> on<\/span> January 4<\/span>th.\u001b[0<\/span>m<\/span>\n\u001b[1<\/span>m><\/span> Finished chain.\u001b[0<\/span>m<\/span>\n<\/span>\nObservation: \u001b[33<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m The stock prices of<\/span> 'ABC'<\/span> and<\/span> 'XYZ'<\/span> on<\/span> January 3<\/span>rd and<\/span> January 4<\/span>th were 232.0<\/span> and<\/span> 798.0<\/span> respectively for<\/span> 'ABC'<\/span> and<\/span> 'XYZ'<\/span> on<\/span> January 3<\/span>rd, and<\/span> 225.0<\/span> and<\/span> 795.0<\/span> respectively for<\/span> 'ABC'<\/span> and<\/span> 'XYZ'<\/span> on<\/span> January 4<\/span>th.\u001b[0<\/span>m<\/span>\nThought:\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I need to<\/span> calculate the ratio between<\/span> the two stock prices on<\/span> each<\/span> day<\/span><\/span>\nAction: Calculator<\/span>\nAction Input: 232.0<\/span>\/<\/span>798.0<\/span> and<\/span> 225.0<\/span>\/<\/span>795.0<\/span>\u001b[0<\/span>m<\/span>\nObservation: \u001b[36<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>mAnswer: 0.2907268170426065<\/span><\/span>\n0.2830188679245283<\/span><\/span>\n\u001b[0<\/span>m<\/span>\nThought:\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m I need to<\/span> calculate the multiplication of<\/span> the two ratios<\/span>\nAction: Calculator<\/span>\nAction Input: 0.2907268170426065<\/span> *<\/span> 0.2830188679245283<\/span>\u001b[0<\/span>m<\/span>\nObservation: \u001b[36<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>mAnswer: 0.08228117463469994<\/span><\/span>\n\u001b[0<\/span>m<\/span>\nThought:\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m><\/span> Finished chain.\u001b[0<\/span>m<\/span>\n<\/code><\/pre>\n
    \u6211\u4eec\u53ef\u4ee5\u770b\u5230\u5f88\u591a\u8f93\u51fa\u3002\u5728\u6bcf\u4e2a\u6b65\u9aa4\u4e2d\uff0c\u90fd\u6709\u4e00\u4e2aThought\uff0c\u5bfc\u81f4\u9009\u62e9\u7684Action\u548cAction Input\u3002\u5982\u679cAction\u8981\u4f7f\u7528\u5de5\u5177\uff0c\u90a3\u4e48\u89c2\u5bdf\uff08\u5de5\u5177\u7684\u8f93\u51fa\uff09\u4f1a\u4f20\u9012\u56de\u4ee3\u7406\u4eba\u3002<\/p>\n
    \u5982\u679c\u6211\u4eec\u770b\u4e00\u4e0b\u4ee3\u7406\u4eba\u4f7f\u7528\u7684\u63d0\u793a\uff0c\u6211\u4eec\u53ef\u4ee5\u770b\u5230LLM\u5982\u4f55\u51b3\u5b9a\u4f7f\u7528\u54ea\u4e2a\u5de5\u5177\u3002<\/p>\n
    In[17]:<\/p>\n
    <\/div>
    scss<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>print<\/span>(zero_shot_agent.agent.llm_chain.prompt.template)<\/span>\n<\/code><\/pre>\n
    Out[17]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>Answer the following questions as<\/span> best you can. You have access to<\/span> the following tools:<\/span>\n<\/span>\nCalculator:<\/span> Useful for<\/span> when<\/span> you need to<\/span> answer questions about math.<\/span>\nStock DB: Useful for<\/span> when<\/span> you need to<\/span> answer questions about stocks and<\/span> their prices.<\/span>\n<\/span>\nUse the following format:<\/span>\n<\/span>\nQuestion:<\/span> the input question you must answer<\/span>\nThought:<\/span> you should always think about what to<\/span> do<\/span><\/span>\nAction:<\/span> the action to<\/span> take<\/span>, should be one of<\/span> [Calculator, Stock DB]<\/span>\nAction Input: the input to<\/span> the action<\/span>\nObservation:<\/span> the result of<\/span> the action<\/span>\n... (this Thought\/Action\/Action Input\/Observation can repeat N times)<\/span>\nThought:<\/span> I now know the final answer<\/span>\nFinal Answer: the final answer to<\/span> the original input question<\/span>\n<\/span>\nBegin!<\/span>\n<\/span>\nQuestion:<\/span> {input}<\/span>\nThought:<\/span>{agent_scratchpad}<\/span>\n<\/code><\/pre>\n
    \u9996\u5148\uff0c\u6211\u4eec\u544a\u8bc9LLM\u5b83\u53ef\u4ee5\u4f7f\u7528\u7684\u5de5\u5177\uff08\u8ba1\u7b97\u5668\u548c\u80a1\u7968\u6570\u636e\u5e93\uff09\u3002\u5728\u6b64\u4e4b\u540e\uff0c\u5b9a\u4e49\u4e86\u4e00\u4e2a\u793a\u4f8b\u683c\u5f0f\uff1b\u8fd9\u9075\u5faa\u4e86\u95ee\u9898\uff08\u6765\u81ea\u7528\u6237\uff09\u3001\u601d\u8003\u3001\u884c\u52a8\u3001\u884c\u52a8\u8f93\u5165\u3001\u89c2\u5bdf\u7684\u6d41\u7a0b – \u5e76\u4e14\u91cd\u590d\u76f4\u5230\u8fbe\u5230\u6700\u7ec8\u7b54\u6848\u3002<\/p>\n
    \u8fd9\u4e9b\u5de5\u5177\u548c\u601d\u7ef4\u8fc7\u7a0b\u5c06\u4ee3\u7406\u4eba\u4e0eLangChain\u4e2d\u7684\u94fe\u5206\u5f00\u3002<\/p>\n
    \u800c\u94fe\u5b9a\u4e49\u4e86\u4e00\u4e2a\u76f4\u63a5\u7684\u8f93\u5165\/\u8f93\u51fa\u8fc7\u7a0b\uff0c\u4ee3\u7406\u4eba\u7684\u903b\u8f91\u5141\u8bb8\u9010\u6b65\u7684\u601d\u8003\u8fc7\u7a0b\u3002\u8fd9\u4e2a\u9010\u6b65\u7684\u8fc7\u7a0b\u7684\u4f18\u52bf\u5728\u4e8eLLM\u53ef\u4ee5\u901a\u8fc7\u591a\u4e2a\u63a8\u7406\u6b65\u9aa4\u6216\u5de5\u5177\u6765\u4ea7\u751f\u66f4\u597d\u7684\u7b54\u6848\u3002<\/p>\n
    \u6211\u4eec\u4ecd\u7136\u9700\u8981\u8ba8\u8bba\u63d0\u793a\u7684\u4e00\u90e8\u5206\u3002\u6700\u540e\u4e00\u884c\u662f\u201cThought:{agent_scratchpad}\u201d\u3002<\/p>\n
    agent_scratchpad\u662f\u6211\u4eec\u6dfb\u52a0\u4ee3\u7406\u4eba\u5df2\u7ecf\u6267\u884c\u7684\u6bcf\u4e2a\u601d\u8003\u6216\u884c\u52a8\u7684\u5730\u65b9\u3002\u6240\u6709\u7684\u601d\u8003\u548c\u884c\u52a8\uff08\u5728\u5f53\u524d\u4ee3\u7406\u4eba\u6267\u884c\u94fe\u5185\uff09\u90fd\u53ef\u4ee5\u88ab\u4e0b\u4e00\u4e2a\u601d\u8003-\u884c\u52a8-\u89c2\u5bdf\u5faa\u73af\u8bbf\u95ee\uff0c\u4ece\u800c\u5b9e\u73b0\u4ee3\u7406\u4eba\u884c\u52a8\u7684\u8fde\u7eed\u6027\u3002<\/p>\n
    \u4f1a\u8bdd\u5f0f\u53cd\u5e94 (Conversational ReAct)<\/h2>\n
    \u96f6\u6b21\u53cd\u5e94\u4ee3\u7406\u4eba\u6548\u679c\u826f\u597d\uff0c\u4f46\u7f3a\u4e4f\u4f1a\u8bdd\u8bb0\u5fc6\u3002\u8fd9\u79cd\u7f3a\u4e4f\u8bb0\u5fc6\u5bf9\u4e8e\u9700\u8981\u8bb0\u4f4f\u5bf9\u8bdd\u4e2d\u5148\u524d\u4ea4\u4e92\u7684\u804a\u5929\u673a\u5668\u4eba\u7c7b\u578b\u7684\u7528\u4f8b\u53ef\u80fd\u4f1a\u6709\u95ee\u9898\u3002<\/p>\n
    \u5e78\u8fd0\u7684\u662f\uff0c\u6211\u4eec\u53ef\u4ee5\u4f7f\u7528\u4f1a\u8bdd\u5f0f\u53cd\u5e94\u63cf\u8ff0\u4ee3\u7406\u4eba\u6765\u8bb0\u4f4f\u4ea4\u4e92\u3002\u6211\u4eec\u53ef\u4ee5\u5c06\u8fd9\u4e2a\u4ee3\u7406\u4eba\u770b\u4f5c\u662f\u4e0e\u6211\u4eec\u4e4b\u524d\u7684\u96f6\u6b21\u53cd\u5e94\u4ee3\u7406\u4eba\u76f8\u540c\uff0c\u4f46\u5177\u6709\u4f1a\u8bdd\u8bb0\u5fc6\u3002<\/p>\n
    \u8981\u521d\u59cb\u5316\u4ee3\u7406\u4eba\uff0c\u6211\u4eec\u9996\u5148\u9700\u8981\u521d\u59cb\u5316\u6211\u4eec\u60f3\u8981\u4f7f\u7528\u7684\u8bb0\u5fc6\u3002\u6211\u4eec\u5c06\u4f7f\u7528\u7b80\u5355\u7684ConversationBufferMemory\u3002<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain.memory import ConversationBufferMemory<\/span>\n<\/span>\nmemory<\/span> = ConversationBufferMemory(memory_key=\"chat_history\"<\/span>)<\/span>\n<\/code><\/pre>\n
    \u6211\u4eec\u5c06\u8fd9\u4e2a\u4f20\u9012\u7ed9\u521d\u59cb\u5316\u4ee3\u7406\u4eba\u65f6\u7684memory\u53c2\u6570\uff1a<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>conversational_agent<\/span> = initialize_agent(<\/span>\n    agent<\/span>='conversational-react-description'<\/span>, <\/span>\n    tools<\/span>=tools, <\/span>\n    llm<\/span>=llm,<\/span>\n    verbose<\/span>=True<\/span>,<\/span>\n    max_iterations<\/span>=3<\/span>,<\/span>\n    memory<\/span>=memory,<\/span>\n)<\/span>\n<\/code><\/pre>\n
    \u5982\u679c\u6211\u4eec\u4f7f\u7528\u7c7b\u4f3c\u7684\u95ee\u9898\u8fd0\u884c\u8fd9\u4e2a\u4ee3\u7406\u4eba\uff0c\u6211\u4eec\u5e94\u8be5\u4f1a\u770b\u5230\u4e0e\u4e4b\u524d\u76f8\u4f3c\u7684\u8fc7\u7a0b\uff1a<\/p>\n
    In[22]:<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>result<\/span> = conversational_agent(<\/span>\n    \"Please provide me the stock prices for ABC on January the 1st\"<\/span>\n)<\/span>\n<\/code><\/pre>\n
    Out[22]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> AgentExecutor chain...\u001b[0<\/span>m<\/span>\n\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m<\/span>\nThought:<\/span> Do<\/span> I need to<\/span> use a tool? Yes<\/span>\nAction:<\/span> Stock DB<\/span>\nAction Input: ABC on<\/span> January the 1s<\/span>t\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> SQLDatabaseChain chain...\u001b[0<\/span>m<\/span>\nABC on<\/span> January the 1s<\/span>t <\/span>\nSQLQuery:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m SELECT<\/span> price FROM<\/span> stocks WHERE<\/span> stock_ticker = 'ABC' AND date = '2023-01-01'\u001b[0m<\/span><\/span>\nSQLResult:<\/span> \u001b[33<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m[(200.0<\/span>,)]\u001b[0<\/span>m<\/span>\nAnswer:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m The price of<\/span> ABC on<\/span> January the 1s<\/span>t was 200.0<\/span>.\u001b[0<\/span>m<\/span>\n\u001b[1<\/span>m> Finished chain.\u001b[0<\/span>m<\/span>\n<\/span>\nObservation:<\/span> \u001b[33<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m The price of<\/span> ABC on<\/span> January the 1s<\/span>t was 200.0<\/span>.\u001b[0<\/span>m<\/span>\nThought:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m Do<\/span> I need to<\/span> use a tool? No<\/span>\nAI:<\/span> Is<\/span> there anything else<\/span> I can help you with<\/span>?\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m> Finished chain.\u001b[0<\/span>m<\/span>\n<\/code><\/pre>\n
    \u5230\u76ee\u524d\u4e3a\u6b62\uff0c\u8fd9\u770b\u8d77\u6765\u4e0e\u6211\u4eec\u4e0a\u4e00\u4e2a\u96f6\u6b21\u53cd\u5e94\u4ee3\u7406\u4eba\u975e\u5e38\u76f8\u4f3c\u3002\u4f46\u4e0e\u6211\u4eec\u7684\u96f6\u6b21\u53cd\u5e94\u4ee3\u7406\u4eba\u4e0d\u540c\uff0c\u6211\u4eec\u73b0\u5728\u53ef\u4ee5\u63d0\u51fa\u540e\u7eed\u95ee\u9898\u3002\u8ba9\u6211\u4eec\u8be2\u95eeXYZ\u5728\u540c\u4e00\u65e5\u671f\u7684\u80a1\u4ef7\uff0c\u800c\u4e0d\u6307\u5b9a1\u67081\u65e5\u3002<\/p>\n
    In[24]:<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>result<\/span> = conversational_agent(<\/span>\n    \"What are the stock prices for XYZ on the same day?\"<\/span>\n)<\/span>\n<\/code><\/pre>\n
    Out[24]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> AgentExecutor chain...\u001b[0<\/span>m<\/span>\n\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m<\/span>\nThought:<\/span> Do<\/span> I need to<\/span> use a tool? Yes<\/span>\nAction:<\/span> Stock DB<\/span>\nAction Input: Stock prices for<\/span> XYZ on<\/span> January 1s<\/span>t\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> SQLDatabaseChain chain...\u001b[0<\/span>m<\/span>\nStock prices for<\/span> XYZ on<\/span> January 1s<\/span>t <\/span>\nSQLQuery:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m SELECT<\/span> price FROM<\/span> stocks WHERE<\/span> stock_ticker = 'XYZ' AND date = '2023-01-01'\u001b[0m<\/span><\/span>\nSQLResult:<\/span> \u001b[33<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m[(810.0<\/span>,)]\u001b[0<\/span>m<\/span>\nAnswer:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m The stock price for<\/span> XYZ on<\/span> January 1s<\/span>t was 810.0<\/span>.\u001b[0<\/span>m<\/span>\n\u001b[1<\/span>m> Finished chain.\u001b[0<\/span>m<\/span>\n<\/span>\nObservation:<\/span> \u001b[33<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m The stock price for<\/span> XYZ on<\/span> January 1s<\/span>t was 810.0<\/span>.\u001b[0<\/span>m<\/span>\nThought:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m Do<\/span> I need to<\/span> use a tool? No<\/span>\nAI:<\/span> Is<\/span> there anything else<\/span> I can help you with<\/span>?\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m> Finished chain.\u001b[0<\/span>m<\/span>\n<\/code><\/pre>\n
    \u6211\u4eec\u53ef\u4ee5\u5728\u7b2c\u4e00\u4e2aAction Input\u4e2d\u770b\u5230\uff0c\u4ee3\u7406\u4eba\u6b63\u5728\u5bfb\u627e\u201c1\u67081\u65e5XYZ\u7684\u80a1\u4ef7\u201d\u3002\u5b83\u77e5\u9053\u6211\u4eec\u8981\u627e1\u67081\u65e5\u7684\u6570\u636e\uff0c\u56e0\u4e3a\u6211\u4eec\u5728\u5148\u524d\u7684\u4ea4\u4e92\u4e2d\u8981\u6c42\u4e86\u8fd9\u4e2a\u65e5\u671f\u3002<\/p>\n
    \u5b83\u662f\u5982\u4f55\u505a\u5230\u8fd9\u4e00\u70b9\u7684\u5462\uff1f\u6211\u4eec\u53ef\u4ee5\u67e5\u770b\u63d0\u793a\u6a21\u677f\u6765\u627e\u51fa\u7b54\u6848\uff1a<\/p>\n
    In[23]:<\/p>\n
    <\/div>
    scss<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>print<\/span>(conversational_agent.agent.llm_chain.prompt.template)<\/span>\n<\/code><\/pre>\n
    Out[23]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>Assistant is<\/span> a large language model trained by<\/span> OpenAI.<\/span>\n<\/span>\nAssistant is<\/span> designed to<\/span> be able to<\/span> assist with<\/span> a wide range of<\/span> tasks, from<\/span> answering simple questions to<\/span> providing in<\/span>-depth explanations and<\/span> discussions on<\/span> a wide range of<\/span> topics. As<\/span> a language model, Assistant is<\/span> able to<\/span> generate human-like<\/span> text<\/span> based on<\/span> the input it receives, allowing it to<\/span> engage in<\/span> natural-sounding conversations and<\/span> provide responses that are coherent and<\/span> relevant to<\/span> the topic at hand.<\/span>\n<\/span>\nAssistant is<\/span> constantly learning and<\/span> improving, and<\/span> its capabilities are constantly evolving. It is<\/span> able to<\/span> process and<\/span> understand large amounts of<\/span> text<\/span>, and<\/span> can use this knowledge to<\/span> provide accurate and<\/span> informative responses to<\/span> a wide range of<\/span> questions. Additionally, Assistant is<\/span> able to<\/span> generate its own text<\/span> based on<\/span> the input it receives, allowing it to<\/span> engage in<\/span> discussions and<\/span> provide explanations and<\/span> descriptions on<\/span> a wide range of<\/span> topics.<\/span>\n<\/span>\nOverall, Assistant is<\/span> a powerful tool that can help with<\/span> a wide range of<\/span> tasks and<\/span> provide valuable insights and<\/span> information on<\/span> a wide range of<\/span> topics. Whether you need help with<\/span> a specific question or<\/span> just want to<\/span> have a conversation about a particular topic, Assistant is<\/span> here to<\/span> assist.<\/span>\n<\/span>\nTOOLS:<\/span><\/span>\n------<\/span>\n<\/span>\nAssistant has access to<\/span> the following tools:<\/span>\n<\/span>\n> Calculator: Useful for<\/span> when<\/span> you need to<\/span> answer questions about math.<\/span>\n> Stock DB: Useful for<\/span> when<\/span> you need to<\/span> answer questions about stocks and<\/span> their prices.<\/span>\n<\/span>\nTo<\/span> use a tool, please use the following format:<\/span>\n<\/span>\n```<\/span>\nThought:<\/span> Do<\/span> I need to<\/span> use a tool? Yes<\/span>\nAction:<\/span> the action to<\/span> take<\/span>, should be one of<\/span> [Calculator, Stock DB]<\/span>\nAction Input: the input to<\/span> the action<\/span>\nObservation:<\/span> the result of<\/span> the action<\/span>\n```<\/span>\n<\/span>\nWhen<\/span> you have a response to<\/span> say to<\/span> the Human, or<\/span> if<\/span> you do<\/span> not<\/span> need to<\/span> use a tool, you MUST use the format:<\/span>\n<\/span>\n```<\/span>\nThought:<\/span> Do<\/span> I need to<\/span> use a tool? No<\/span>\nAI:<\/span> [your response here]<\/span>\n```<\/span>\n<\/span>\nBegin!<\/span>\n<\/span>\nPrevious conversation history:<\/span>\n{chat_history}<\/span>\n<\/span>\nNew<\/span> input: {input}<\/span>\n{agent_scratchpad}<\/span>\n<\/code><\/pre>\n
    \u5728\u63d0\u793a\u7684\u5f00\u5934\uff0c\u6211\u4eec\u6709\u4e00\u4e2a\u66f4\u5927\u7684\u6307\u4ee4\u8bbe\u7f6e\uff0c\u4f46\u6700\u91cd\u8981\u7684\u662f\u63d0\u793a\u672b\u5c3e\u9644\u8fd1\u7684\u4e24\u884c\uff1a<\/p>\n
    \u4ee5\u524d\u7684\u5bf9\u8bdd\u5386\u53f2\uff1a{chat_history}<\/p>\n
    \u5728\u8fd9\u91cc\uff0c\u6211\u4eec\u5c06\u6240\u6709\u5148\u524d\u7684\u4ea4\u4e92\u6dfb\u52a0\u5230\u63d0\u793a\u4e2d\u3002\u5728\u8fd9\u4e2a\u7a7a\u95f4\u4e2d\u5c06\u5305\u542b\u6211\u4eec\u8be2\u95ee\u7684\u201c\u8bf7\u63d0\u4f9bABC\u80a1\u7968\u57281\u67081\u65e5\u7684\u80a1\u4ef7\u201d\u7684\u4fe1\u606f – \u8fd9\u4f7f\u4ee3\u7406\u4eba\u80fd\u591f\u7406\u89e3\u6211\u4eec\u7684\u540e\u7eed\u95ee\u9898\u6307\u7684\u662f\u76f8\u540c\u7684\u65e5\u671f\u3002<\/p>\n
    \u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u4f1a\u8bdd\u5f0fReAct\u4ee3\u7406\u4eba\u4e13\u4e3a\u5bf9\u8bdd\u8bbe\u8ba1\uff0c\u5f53\u7ed3\u5408\u591a\u4e2a\u590d\u6742\u6b65\u9aa4\u65f6\uff0c\u5b83\u6bd4\u96f6\u6b21\u53cd\u5e94\u4ee3\u7406\u4eba\u66f4\u5bb9\u6613\u9047\u5230\u56f0\u96be\u3002\u5982\u679c\u6211\u4eec\u8981\u6c42\u4ee3\u7406\u4eba\u56de\u7b54\u6211\u4eec\u4e4b\u524d\u7684\u95ee\u9898\uff0c\u6211\u4eec\u53ef\u4ee5\u770b\u5230\u8fd9\u4e00\u70b9\uff1a<\/p>\n
    In[26]:<\/p>\n
    <\/div>
    makefile<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>result = conversational_agent(<\/span>\n    \"What is the multiplication of the ratio of the prices of stocks 'ABC' \"<\/span><\/span>\n    \"and 'XYZ' in January 3rd and the ratio of the same prices of the same \"<\/span><\/span>\n    \"stocks in January the 4th?\"<\/span><\/span>\n)<\/span>\n<\/code><\/pre>\n
    Out[26]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> AgentExecutor chain...\u001b[0<\/span>m<\/span>\n\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>mThought: Do<\/span> I need to<\/span> use a tool? Yes<\/span>\nAction:<\/span> Stock DB<\/span>\nAction Input: Get<\/span> the ratio of<\/span> the prices of<\/span> stocks 'ABC' and 'XYZ' in January 3rd and the ratio of the same prices of the same stocks in January the 4th\u001b[0m<\/span><\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> SQLDatabaseChain chain...\u001b[0<\/span>m<\/span>\nGet<\/span> the ratio of<\/span> the prices of<\/span> stocks 'ABC' and 'XYZ' in January 3rd and the ratio of the same prices of the same stocks in January the 4th <\/span><\/span>\nSQLQuery:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m SELECT<\/span> (SELECT<\/span> price FROM<\/span> stocks WHERE<\/span> stock_ticker = 'ABC' AND date = '2023-01-03') \/ (SELECT price FROM stocks WHERE stock_ticker = 'XYZ' AND date = '2023-01-03') AS ratio_jan_3, (SELECT price FROM stocks WHERE stock_ticker = 'ABC' AND date = '2023-01-04') \/ (SELECT price FROM stocks WHERE stock_ticker = 'XYZ' AND date = '2023-01-04') AS ratio_jan_4 FROM stocks LIMIT 5;\u001b[0m<\/span><\/span>\nSQLResult:<\/span> \u001b[33<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m[(0.2907268170426065<\/span>, 0.2830188679245283<\/span>), (0.2907268170426065<\/span>, 0.2830188679245283<\/span>), (0.2907268170426065<\/span>, 0.2830188679245283<\/span>), (0.2907268170426065<\/span>, 0.2830188679245283<\/span>), (0.2907268170426065<\/span>, 0.2830188679245283<\/span>)]\u001b[0<\/span>m<\/span>\nAnswer:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m The ratio of<\/span> the prices of<\/span> stocks 'ABC' and 'XYZ' in January 3rd is 0.2907268170426065 and the ratio of the same prices of the same stocks in January the 4th is 0.2830188679245283.\u001b[0m<\/span><\/span>\n\u001b[1<\/span>m> Finished chain.\u001b[0<\/span>m<\/span>\n<\/span>\nObservation:<\/span> \u001b[33<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m The ratio of<\/span> the prices of<\/span> stocks 'ABC' and 'XYZ' in January 3rd is 0.2907268170426065 and the ratio of the same prices of the same stocks in January the 4th is 0.2830188679245283.\u001b[0m<\/span><\/span>\nThought:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m Do<\/span> I need to<\/span> use a tool? No<\/span>\nAI:<\/span> The answer is<\/span> 0.4444444444444444<\/span>. Is<\/span> there anything else<\/span> I can help you with<\/span>?\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m> Finished chain.\u001b[0<\/span>m<\/span>\nSpent a total of<\/span> 2518<\/span> tokens<\/span>\n<\/code><\/pre>\n
    \u6709\u4e86\u8fd9\u4e2a\uff0c\u4ee3\u7406\u4eba\u4ecd\u7136\u80fd\u591f\u89e3\u51b3\u95ee\u9898\uff0c\u4f46\u4f7f\u7528\u4e86\u66f4\u590d\u6742\u7684\u7eafSQL\u65b9\u6cd5\uff0c\u800c\u4e0d\u662f\u4f9d\u8d56\u66f4\u7b80\u5355\u7684SQL\u548c\u8ba1\u7b97\u5668\u5de5\u5177\u3002<\/p>\n
    ReAct\u6587\u6863\u5b58\u50a8 (ReAct Docstore)<\/h2>\n
    \u53e6\u4e00\u4e2a\u5e38\u89c1\u7684\u4ee3\u7406\u662freact-docstore\u4ee3\u7406\u3002\u4e0e\u4e4b\u524d\u4e00\u6837\uff0c\u5b83\u4f7f\u7528ReAct\u65b9\u6cd5\uff0c\u4f46\u73b0\u5728\u660e\u786e\u5730\u6784\u5efa\u7528\u4e8e\u4fe1\u606f\u641c\u7d22\u548c\u67e5\u627e\u7684LangChain\u6587\u6863\u5b58\u50a8\u3002<\/p>\n
    LangChain\u6587\u6863\u5b58\u50a8\u5141\u8bb8\u6211\u4eec\u4f7f\u7528\u4f20\u7edf\u7684\u68c0\u7d22\u65b9\u6cd5\u5b58\u50a8\u548c\u68c0\u7d22\u4fe1\u606f\u3002\u5176\u4e2d\u4e00\u4e2a\u6587\u6863\u5b58\u50a8\u662f\u7ef4\u57fa\u767e\u79d1\uff0c\u5b83\u8ba9\u6211\u4eec\u8bbf\u95ee\u8be5\u7ad9\u70b9\u4e0a\u7684\u4fe1\u606f\u3002<\/p>\n
    \u6211\u4eec\u5c06\u4f7f\u7528\u4e24\u79cd\u6587\u6863\u5b58\u50a8\u65b9\u6cd5\u6765\u5b9e\u73b0\u6b64\u4ee3\u7406 \u2014 \u641c\u7d22\u548c\u67e5\u627e\u3002\u4f7f\u7528\u641c\u7d22\uff0c\u6211\u4eec\u7684\u4ee3\u7406\u5c06\u641c\u7d22\u76f8\u5173\u7684\u6587\u7ae0\uff0c\u800c\u4f7f\u7528\u67e5\u627e\uff0c\u4ee3\u7406\u5c06\u5728\u68c0\u7d22\u5230\u7684\u6587\u7ae0\u4e2d\u627e\u5230\u76f8\u5173\u7684\u4fe1\u606f\u5757\u3002\u8981\u521d\u59cb\u5316\u8fd9\u4e24\u4e2a\u5de5\u5177\uff0c\u6211\u4eec\u6267\u884c\uff1a<\/p>\n
    <\/div>
    go<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain import<\/span> Wikipedia<\/span>\nfrom langchain.agents.react.base import<\/span> DocstoreExplorer<\/span>\n<\/span>\ndocstore=DocstoreExplorer(Wikipedia())<\/span>\ntools = [<\/span>\n    Tool(<\/span>\n        name=\"Search\"<\/span>,<\/span>\n        func<\/span>=docstore<\/span>.search<\/span>,<\/span><\/span>\n        description='search wikipedia'<\/span><\/span>\n    ),<\/span>\n    Tool(<\/span>\n        name=\"Lookup\"<\/span>,<\/span>\n        func<\/span>=docstore<\/span>.lookup<\/span>,<\/span><\/span>\n        description='lookup a term in wikipedia'<\/span><\/span>\n    )<\/span>\n]<\/span>\n<\/code><\/pre>\n
    \u73b0\u5728\u521d\u59cb\u5316\u4ee3\u7406\uff1a<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>docstore_agent<\/span> = initialize_agent(<\/span>\n    tools, <\/span>\n    llm, <\/span>\n    agent<\/span>=\"react-docstore\"<\/span>, <\/span>\n    verbose<\/span>=True<\/span>,<\/span>\n    max_iterations<\/span>=3<\/span><\/span>\n)<\/span>\n<\/code><\/pre>\n
    \u8ba9\u6211\u4eec\u8bd5\u4e00\u4e0b\u4e0b\u9762\u7684\uff1a<\/p>\n
    In[30]:<\/p>\n
    <\/div>
    scss<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>docstore_agent<\/span>(\"What were Archimedes' last words?\")<\/span>\n<\/code><\/pre>\n
    Out[30]:<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\n\u001b[1<\/span>m> Entering new<\/span> AgentExecutor chain...\u001b[0<\/span>m<\/span>\n\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>mThought: I need to<\/span> search Archimedes and<\/span> find his last words.<\/span>\nAction:<\/span> Search[Archimedes]\u001b[0<\/span>m<\/span>\nObservation:<\/span> \u001b[36<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>mArchimedes of<\/span> Syracuse (; c.\u2009287<\/span> \u2013 c.\u2009212<\/span> BC) was a Greek mathematician, physicist, engineer...\u001b[0<\/span>m<\/span>\nThought:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m The paragraph does not<\/span> mention Archimedes' last words. I need to look up \"last words\".<\/span><\/span>\nAction:<\/span> Lookup[last words]\u001b[0<\/span>m<\/span>\nObservation:<\/span> \u001b[33<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m(Result 1<\/span>\/1<\/span>) Plutarch (45<\/span>\u2013119<\/span> AD) wrote in<\/span> his Parallel Lives that Archimedes was related to<\/span> King Hiero...\u001b[0<\/span>m<\/span>\nThought:<\/span>\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m The last words attributed to<\/span> Archimedes are \"Do not disturb my circles\"<\/span>, so the answer is<\/span> \"Do not disturb my circles\"<\/span>.<\/span>\nAction:<\/span> Finish[\"Do not disturb my circles\"<\/span>]\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m> Finished chain.\u001b[0<\/span>m<\/span>\n<\/code><\/pre>\n
    Out[30]:<\/p>\n
    <\/div>
    css<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>{'input<\/span>': \"What were Archimedes' last words?\"<\/span>,<\/span>\n 'output'<\/span>: '\"Do not disturb my circles\"'<\/span>}<\/span>\n<\/code><\/pre>\n
    \u8fd9\u4e2a\u4ee3\u7406\u7684\u63d0\u793a\u975e\u5e38\u957f\uff0c\u56e0\u6b64\u6211\u4eec\u5c06\u5c55\u793a\u4e00\u4e2a\u7f29\u77ed\u7248\u672c\u3002\u5b83\u53ea\u5305\u542b\u4e86\u8fd9\u4e2a\u4ee3\u7406\u5e94\u8be5\u5982\u4f55\u4f7f\u7528\u641c\u7d22\u548c\u67e5\u627e\u5de5\u5177\u7684\u82e5\u5e72\u793a\u4f8b\uff1a<\/p>\n
    <\/div>
    vbnet<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>Question:<\/span> What is<\/span> the elevation range for<\/span> the area that the eastern sector of<\/span> the Colorado orogeny extends into<\/span>?<\/span>\n<\/span>\nThought:<\/span> I need to<\/span> search Colorado orogeny, find the area that the eastern sector of<\/span> the Colorado orogeny extends into<\/span>, then<\/span> find the elevation range of<\/span> the area.<\/span>\nAction:<\/span> Search[Colorado orogeny]<\/span>\nObservation:<\/span> The Colorado orogeny was an episode of<\/span> mountain building (an orogeny) in<\/span> Colorado and<\/span> surrounding areas.<\/span>\n<\/span>\nThought:<\/span> It does not<\/span> mention the eastern sector. So I need to<\/span> look up eastern sector.<\/span>\nAction:<\/span> Lookup[eastern sector]<\/span>\nObservation:<\/span> (Result 1<\/span> \/ 1<\/span>) The eastern sector extends into<\/span> the High Plains and<\/span> is<\/span> called the Central Plains orogeny.<\/span>\n<\/span>\nThought:<\/span> The eastern sector of<\/span> Colorado orogeny extends into<\/span> the High Plains. So I need to<\/span> search High Plains and<\/span> find its elevation range.<\/span>\nAction:<\/span> Search[High Plains]<\/span>\nObservation:<\/span> High Plains refers to<\/span> one of<\/span> two distinct<\/span> land regions<\/span>\n<\/span>\nThought:<\/span> I need to<\/span> instead search High Plains (United States).<\/span>\nAction:<\/span> Search[High Plains (United States)]<\/span>\nObservation:<\/span> The High Plains are a subregion of<\/span> the Great Plains. From<\/span> east to<\/span> west, the High Plains rise in<\/span> elevation from<\/span> around 1<\/span>,800<\/span> to<\/span> 7<\/span>,000<\/span> ft (550<\/span> to<\/span> 2<\/span>,130<\/span> m).[3<\/span>]<\/span>\n<\/span>\nThought:<\/span> High Plains rise in<\/span> elevation from<\/span> around 1<\/span>,800<\/span> to<\/span> 7<\/span>,000<\/span> ft, so the answer is<\/span> 1<\/span>,800<\/span> to<\/span> 7<\/span>,000<\/span> ft.<\/span>\nAction:<\/span> Finish[1<\/span>,800<\/span> to<\/span> 7<\/span>,000<\/span> ft]<\/span>\n<\/code><\/pre>\n
    \u8fd9\u4e2a\u63d0\u793a\u5305\u542b\u82e5\u5e72\u7c7b\u4f3c\u683c\u5f0f\u7684\u793a\u4f8b\u3002\u5728\u63d0\u793a\u7684\u672b\u5c3e\uff0c\u6211\u4eec\u770b\u5230\uff1a<\/p>\n
    <\/div>
    css<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>Question: {input<\/span>}<\/span>\n{agent_scratchpad}<\/span>\n<\/code><\/pre>\n
    \u4e0e\u4e4b\u524d\u4e00\u6837\uff0c\u6211\u4eec\u6709\u4e00\u4e2a{input}\uff0c\u7528\u6765\u4f20\u9012\u6700\u8fd1\u7684\u7528\u6237\u67e5\u8be2\uff0c\u4ee5\u53ca\u4e00\u4e2a{agent_scratchpad}\uff0c\u7528\u6765\u8ddf\u8e2a\u4e4b\u524d\u7684\u60f3\u6cd5\u548c\u884c\u52a8\u3002<\/p>\n
    \u4e0e\u6211\u4eec\u7684\u5bf9\u8bdd\u4ee3\u7406\u4e0d\u540c\uff0c\u8fd9\u91cc\u6ca1\u6709{chat_history}\u8f93\u5165\u3002\u8fd9\u610f\u5473\u7740\u6211\u4eec\u6b63\u5728\u4f7f\u7528\u53e6\u4e00\u4e2a\u96f6-shot\u4ee3\u7406\u3002<\/p>\n
    \u81ea\u6211\u63d0\u95ee\u4e0e\u641c\u7d22<\/h2>\n
    \u8ba9\u6211\u4eec\u770b\u4e00\u4e0b\u6700\u540e\u4e00\u4e2a\u4ee3\u7406\u2014\u2014\u81ea\u6211\u63d0\u95ee\u4e0e\u641c\u7d22\u4ee3\u7406\u3002\u5f53\u8fde\u63a5\u4e00\u4e2aLLM\u4e0e\u641c\u7d22\u5f15\u64ce\u65f6\uff0c\u8fd9\u662f\u4f60\u5e94\u8be5\u8003\u8651\u7684\u9996\u9009\u4ee3\u7406\u3002<\/p>\n
    \u8fd9\u4e2a\u4ee3\u7406\u5c06\u6839\u636e\u9700\u8981\u6267\u884c\u641c\u7d22\u5e76\u63d0\u51fa\u8ddf\u8fdb\u95ee\u9898\uff0c\u4ee5\u83b7\u53d6\u6700\u7ec8\u7b54\u6848\u3002\u6211\u4eec\u521d\u59cb\u5316\u8fd9\u4e2a\u4ee3\u7406\u5982\u4e0b\u6240\u793a\uff1a<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain import SerpAPIWrapper<\/span>\n<\/span>\n# initialize the search chain<\/span><\/span>\nsearch<\/span> = SerpAPIWrapper(serpapi_api_key='serp_api_key'<\/span>)<\/span>\n<\/span>\n# create a search tool<\/span><\/span>\ntools<\/span> = [<\/span>\n    Tool(<\/span>\n        name=\"Intermediate Answer\"<\/span>,<\/span>\n        func=search.run,<\/span>\n        description='google search'<\/span><\/span>\n    )<\/span>\n]<\/span>\n<\/span>\n# initialize the search enabled agent<\/span><\/span>\nself_ask_with_search<\/span> = initialize_agent(<\/span>\n    tools,<\/span>\n    llm,<\/span>\n    agent<\/span>=\"self-ask-with-search\"<\/span>,<\/span>\n    verbose<\/span>=True<\/span><\/span>\n)<\/span>\n<\/code><\/pre>\n
    \u73b0\u5728\u8ba9\u6211\u4eec\u63d0\u51fa\u4e00\u4e2a\u9700\u8981\u591a\u6b21\u641c\u7d22\u548c\u201c\u81ea\u6211\u63d0\u95ee\u201d\u6b65\u9aa4\u7684\u95ee\u9898\u3002<\/p>\n
    In[38]:<\/p>\n
    <\/div>
    scss<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>self_ask_with_search<\/span>(<\/span>\n    \"who lived longer; Plato, Socrates, or Aristotle?\"<\/span>\n)<\/span>\n<\/code><\/pre>\n
    Out[38]:<\/p>\n
    <\/div>
    sql<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\n\u001b[1<\/span>m><\/span> Entering new<\/span> AgentExecutor chain...\u001b[0<\/span>m<\/span>\n\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m Yes.<\/span>\nFollow up: How old<\/span> was Plato when<\/span> he died?\u001b[0<\/span>m<\/span>\nIntermediate answer: \u001b[36<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>meighty\u001b[0<\/span>m\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m<\/span>\nFollow up: How old<\/span> was Socrates when<\/span> he died?\u001b[0<\/span>m<\/span>\nIntermediate answer: \u001b[36<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>mapproximately 71<\/span>\u001b[0<\/span>m\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m<\/span>\nFollow up: How old<\/span> was Aristotle when<\/span> he died?\u001b[0<\/span>m<\/span>\nIntermediate answer: \u001b[36<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m62 years\u001b[0<\/span>m\u001b[32<\/span>;1<\/span>m\u001b[1<\/span>;3<\/span>m<\/span>\nSo the final<\/span> answer is<\/span>: Plato\u001b[0<\/span>m<\/span>\n<\/span>\n\u001b[1<\/span>m><\/span> Finished chain.\u001b[0<\/span>m<\/span>\n<\/code><\/pre>\n
    Out[38]:<\/p>\n
    <\/div>
    css<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>{'input<\/span>': 'who lived longer; Plato, Socrates, or Aristotle?'<\/span>,<\/span>\n 'output'<\/span>: 'Plato'<\/span>}<\/span>\n<\/code><\/pre>\n
    \u6211\u4eec\u53ef\u4ee5\u770b\u5230\u4ee3\u7406\u7684\u591a\u6b65\u9aa4\u8fc7\u7a0b\u3002\u5b83\u6267\u884c\u591a\u4e2a\u8ddf\u8fdb\u95ee\u9898\uff0c\u4ee5\u9010\u6e10\u627e\u5230\u6700\u7ec8\u7b54\u6848\u3002<\/p>\n
    \u8fd9\u5c31\u662f\u5173\u4e8eLangChain\u4ee3\u7406\u7684\u8fd9\u4e00\u7ae0\u7684\u5168\u90e8\u5185\u5bb9\u3002\u6b63\u5982\u4f60\u65e0\u7591\u5df2\u7ecf\u6ce8\u610f\u5230\u7684\uff0c\u4ee3\u7406\u5728LangChain\u4e2d\u6db5\u76d6\u4e86\u5e7f\u6cdb\u7684\u5de5\u5177\u8303\u56f4\u3002\u6211\u4eec\u5df2\u7ecf\u6db5\u76d6\u4e86\u8bb8\u591a\u57fa\u672c\u5185\u5bb9\uff0c\u4f46\u8fd8\u6709\u5f88\u591a\u53ef\u4ee5\u8ba8\u8bba\u7684\u5185\u5bb9\u3002<\/p>\n
    \u4ee3\u7406\u7684\u8f6c\u578b\u6f5c\u529b\u5bf9\u4e8e\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u6765\u8bf4\u662f\u4e00\u4e2a\u5de8\u5927\u7684\u98de\u8dc3\uff0c\u800c\u4e14\u5728\u4e0d\u4e45\u7684\u5c06\u6765\uff0c\u201cLLM\u4ee3\u7406\u201d\u8fd9\u4e2a\u672f\u8bed\u5f88\u53ef\u80fd\u4f1a\u4e0eLLM\u672c\u8eab\u53d8\u5f97\u540c\u4e49\u3002<\/p>\n
    \u901a\u8fc7\u8d4b\u4e88LLM\u5229\u7528\u5de5\u5177\u548c\u5728\u8fd9\u4e9b\u4ee3\u7406\u6846\u67b6\u4e2d\u5bfc\u822a\u590d\u6742\u7684\u591a\u6b65\u601d\u7ef4\u8fc7\u7a0b\u7684\u80fd\u529b\uff0c\u6211\u4eec\u6b63\u5728\u6d89\u8db3\u4e00\u4e2a\u5e9e\u5927\u7684\u3001\u7531\u4eba\u5de5\u667a\u80fd\u9a71\u52a8\u7684\u673a\u4f1a\u9886\u57df\u3002<\/p>","protected":false},"excerpt":{"rendered":"
    \u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u975e\u5e38\u5f3a\u5927\uff0c\u4f46\u5b83\u4eec\u7f3a\u4e4f\u4e00\u4e9b\u201c\u6700\u7b28\u201d\u7684\u8ba1\u7b97\u673a\u7a0b\u5e8f\u53ef\u4ee5\u8f7b\u677e\u5904\u7406\u7684\u7279\u5b9a\u80fd\u529b\u3002\u903b\u8f91\u3001\u8ba1\u7b97\u548c\u641c\u7d22\u662f\u8ba1\u7b97\u673a\u901a\u5e38\u64c5\u957f\u7684\u9886\u57df\uff0c\u4f46LLMs\u5219\u8868\u73b0\u4e0d\u4f73\u3002 \u8ba1\u7b97\u673a\u53ef\u4ee5\u89e3\u51b3\u975e\u5e38\u590d\u6742\u7684\u6570\u5b66\u95ee\u9898\uff0c\u7136\u800c\uff0c\u5982\u679c<\/p>\n","protected":false},"author":1,"featured_media":7825,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_title":"","rank_math_description":"","rank_math_focus_keyword":"","views":"6","footnotes":""},"categories":[3],"tags":[136,126,127,128,129],"collection":[],"class_list":["post-1360","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-fenlei2","tag-136","tag-gpt","tag-ai","tag-128","tag-129"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1360","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/comments?post=1360"}],"version-history":[{"count":0,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1360\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media\/7825"}],"wp:attachment":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media?parent=1360"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/categories?post=1360"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/tags?post=1360"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/collection?post=1360"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}