{"id":1967,"date":"2024-05-03T19:18:53","date_gmt":"2024-05-03T19:18:53","guid":{"rendered":"https:\/\/www.nicekj.com\/?p=1967"},"modified":"2024-05-03T19:18:53","modified_gmt":"2024-05-03T19:18:53","slug":"chushilangchaindekuaisurumenzhinan","status":"publish","type":"post","link":"https:\/\/www.nicekj.com\/chushilangchaindekuaisurumenzhinan.html","title":{"rendered":"\u521d\u8bc6LangChain\u7684\u5feb\u901f\u5165\u95e8\u6307\u5357"},"content":{"rendered":"

LangChain<\/h1>\n

\u6982\u8ff0<\/h2>\n
\n

LangChain\u662f\u4e00\u4e2a\u57fa\u4e8e\u5927\u8bed\u8a00\u6a21\u578b\u7528\u4e8e\u6784\u5efa\u7aef\u5230\u7aef\u8bed\u8a00\u6a21\u578b\u5e94\u7528\u7684\u6846\u67b6\uff0c\u5b83\u63d0\u4f9b\u4e86\u4e00\u7cfb\u5217\u5de5\u5177\u3001\u5957\u4ef6\u548c\u63a5\u53e3\uff0c\u8ba9\u5f00\u53d1\u8005\u4f7f\u7528\u8bed\u8a00\u6a21\u578b\u6765\u5b9e\u73b0\u5404\u79cd\u590d\u6742\u7684\u4efb\u52a1\uff0c\u5982\u6587\u672c\u5230\u56fe\u50cf\u7684\u751f\u6210\u3001\u6587\u6863\u95ee\u7b54\u3001\u804a\u5929\u673a\u5668\u4eba\u7b49\u3002<\/p>\n<\/blockquote>\n

LangChain\u7b80\u5316\u4e86LLM\u5e94\u7528\u7a0b\u5e8f\u751f\u547d\u5468\u671f\u7684\u5404\u4e2a\u9636\u6bb5\uff1a<\/p>\n\n

\u5f00\u53d1\u9636\u6bb5\uff1a\u4f7f\u7528LangChain\u7684\u5f00\u6e90\u6784\u5efa\u5757\u548c\u7ec4\u4ef6\u6784\u5efa\u5e94\u7528\u7a0b\u5e8f\uff0c\u5229\u7528\u7b2c\u4e09\u65b9\u96c6\u6210\u548c\u6a21\u677f\u5feb\u901f\u542f\u52a8\u3002\n\n\u751f\u4ea7\u5316\u9636\u6bb5\uff1a\u4f7f\u7528LangSmith\u68c0\u67e5\u3001\u76d1\u63a7\u548c\u8bc4\u4f30\u60a8\u7684\u94fe\uff0c\u4ece\u800c\u53ef\u4ee5\u81ea\u4fe1\u5730\u6301\u7eed\u4f18\u5316\u548c\u90e8\u7f72\u3002\n\n\u90e8\u7f72\u9636\u6bb5\uff1a\u4f7f\u7528LangServe\u5c06\u4efb\u4f55\u94fe\u8f6c\u5316\u4e3aAPI\u3002\n<\/code><\/pre>\n\n
\u53c2\u8003\u6587\u6863\uff1ahttps:\/\/python.langchain.com<\/code><\/p>\n
\u516d\u5927\u7ec4\u4ef6<\/h2>\n
\n
LangChain\u5177\u6709\u516d\u5927\u7ec4\u4ef6,\u5b83\u4eec\u662fLangChain\u7684\u57fa\u77f3\uff0c\u5b83\u4eec\u76f8\u4e92\u534f\u4f5c\uff0c\u5f62\u6210\u4e00\u4e2a\u5f3a\u5927\u800c\u7075\u6d3b\u7684\u7cfb\u7edf\u3002<\/p>\n<\/blockquote>\n\n
\u6a21\u578b\uff08Models\uff09\uff1a\u5305\u542b\u5404\u5927\u8bed\u8a00\u6a21\u578b\u7684LangChain\u63a5\u53e3\u548c\u8c03\u7528\u7ec6\u8282\uff0c\u4ee5\u53ca\u8f93\u51fa\u89e3\u6790\u673a\u5236\u3002\n\n\u63d0\u793a\u6a21\u677f\uff08Prompts\uff09\uff1a\u4f7f\u63d0\u793a\u5de5\u7a0b\u6d41\u7ebf\u5316\uff0c\u8fdb\u4e00\u6b65\u6fc0\u53d1\u5927\u8bed\u8a00\u6a21\u578b\u7684\u6f5c\u529b\u3002\n\n\u6570\u636e\u68c0\u7d22\uff08Indexes\uff09\uff1a\u6784\u5efa\u5e76\u64cd\u4f5c\u6587\u6863\u7684\u65b9\u6cd5\uff0c\u63a5\u53d7\u7528\u6237\u7684\u67e5\u8be2\u5e76\u8fd4\u56de\u6700\u76f8\u5173\u7684\u6587\u6863\uff0c\u8f7b\u677e\u642d\u5efa\u672c\u5730\u77e5\u8bc6\u5e93\u3002\n\n\u8bb0\u5fc6\uff08Memory\uff09\uff1a\u901a\u8fc7\u77ed\u65f6\u8bb0\u5fc6\u548c\u957f\u65f6\u8bb0\u5fc6\uff0c\u5728\u5bf9\u8bdd\u8fc7\u7a0b\u4e2d\u5b58\u50a8\u548c\u68c0\u7d22\u6570\u636e\uff0c\u8ba9ChatBot\u8bb0\u4f4f\u4f60\u3002\n\n\u94fe\uff08Chains\uff09\uff1aLangChain\u4e2d\u7684\u6838\u5fc3\u673a\u5236\uff0c\u4ee5\u7279\u5b9a\u65b9\u5f0f\u5c01\u88c5\u5404\u79cd\u529f\u80fd\uff0c\u5e76\u901a\u8fc7\u4e00\u7cfb\u5217\u7684\u7ec4\u5408\uff0c\u81ea\u52a8\u800c\u7075\u6d3b\u5730\u5b8c\u6210\u4efb\u52a1\u3002\n\n\u4ee3\u7406\uff08Agents\uff09\uff1a\u53e6\u4e00\u4e2aLangChain\u4e2d\u7684\u6838\u5fc3\u673a\u5236\uff0c\u901a\u8fc7\u201c\u4ee3\u7406\u201d\u8ba9\u5927\u6a21\u578b\u81ea\u4e3b\u8c03\u7528\u5916\u90e8\u5de5\u5177\u548c\u5185\u90e8\u5de5\u5177\uff0c\u4f7f\u667a\u80fdAgent\u6210\u4e3a\u53ef\u80fd\u3002\n<\/code><\/pre>\n\n
\u8868\u683c\u5feb\u901f\u9884\u89c8<\/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
\u82f1\u6587<\/th>\u4e2d\u6587<\/th>\u8bf4\u660e<\/th><\/tr><\/thead>
Models<\/td>\u6a21\u578b<\/td>LangChain\u548c\u5927\u6a21\u578b\u7684\u63a5\u53e3<\/td><\/tr>
Chains<\/td>\u94fe<\/td>LangChain\u5bf9\u5404\u79cd\u7ec4\u4ef6\u7684\u5c01\u88c5\u548c\u8c03\u7528\u94fe\u6761<\/td><\/tr>
Prompts<\/td>\u63d0\u793a<\/td>\u8f93\u5165\u5927\u6a21\u578b\u7684\u63d0\u793a\u6a21\u677f<\/td><\/tr>
Indexes<\/td>\u7d22\u5f15<\/td>\u975e\u7ed3\u6784\u5316\u6570\u636e\u7684\u67e5\u627e\u548c\u8bbf\u95ee<\/td><\/tr>
Memory<\/td>\u8bb0\u5fc6\uff08\u5185\u5b58\uff09<\/td>\u4e0e\u5927\u6a21\u578b\u804a\u5929\u8fc7\u7a0b\u4e2d\u8bb0\u5fc6\u4fe1\u606f<\/td><\/tr>
Agents<\/td>\u4ee3\u7406<\/td>\u81ea\u4e3b\u8ba1\u5212\u548c\u6267\u884c\u7684\u673a\u5236<\/td><\/tr><\/tbody><\/table>\n
\u5f00\u6e90\u5e93\u7ec4\u6210<\/h2>\n
LangChain\u6846\u67b6\u7531\u4ee5\u4e0b\u5f00\u6e90\u5e93\u7ec4\u6210\uff1a<\/p>\n\n
langchain-core \uff1a\u57fa\u7840\u62bd\u8c61\u548cLangChain\u8868\u8fbe\u5f0f\u8bed\u8a00\n\nlangchain-community \uff1a\u7b2c\u4e09\u65b9\u96c6\u6210\u3002\u5408\u4f5c\u4f19\u4f34\u5305\uff08\u5982langchain-openai\u3001langchain-anthropic\u7b49\uff09\uff0c\u4e00\u4e9b\u96c6\u6210\u5df2\u7ecf\u8fdb\u4e00\u6b65\u62c6\u5206\u4e3a\u81ea\u5df1\u7684\u8f7b\u91cf\u7ea7\u5305\uff0c\u53ea\u4f9d\u8d56\u4e8elangchain-core\n\nlangchain \uff1a\u6784\u6210\u5e94\u7528\u7a0b\u5e8f\u8ba4\u77e5\u67b6\u6784\u7684\u94fe\u3001\u4ee3\u7406\u548c\u68c0\u7d22\u7b56\u7565\n\nlanggraph\uff1a\u901a\u8fc7\u5c06\u6b65\u9aa4\u5efa\u6a21\u4e3a\u56fe\u4e2d\u7684\u8fb9\u548c\u8282\u70b9\uff0c\u4f7f\u7528 LLMs \u6784\u5efa\u5065\u58ee\u4e14\u6709\u72b6\u6001\u7684\u591a\u53c2\u4e0e\u8005\u5e94\u7528\u7a0b\u5e8f\n\nlangserve\uff1a\u5c06 LangChain \u94fe\u90e8\u7f72\u4e3a REST<\/span> API\n\nLangSmith\uff1a\u4e00\u4e2a\u5f00\u53d1\u8005\u5e73\u53f0\uff0c\u53ef\u8ba9\u60a8\u8c03\u8bd5\u3001\u6d4b\u8bd5\u3001\u8bc4\u4f30\u548c\u76d1\u63a7LLM\u5e94\u7528\u7a0b\u5e8f\uff0c\u5e76\u4e0eLangChain\u65e0\u7f1d\u96c6\u6210\n<\/code><\/pre>\n\n
\n
\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/figure>\n<\/p>\n
LangChain\u57fa\u672c\u4f7f\u7528<\/h1>\n
\u6784\u5efa\u4e00\u4e2a LangChain\u5e94\u7528\u7a0b\u5e8f\u6765\u719f\u6089\u8be5\u6846\u67b6\u3002<\/p>\n
\u5b89\u88c5LangChain<\/h2>\n
\u5b89\u88c5\u6307\u5b9a\u7248\u672c\u7684LangChain\uff0c\u8fd9\u91cc\u5b89\u88c5\u622a\u6b62\u76ee\u524d\u7684\u6700\u65b0\u7248\u672c<\/p>\n\n
pip install langchain==0.1<\/span>.7<\/span>\n<\/code><\/pre>\n\n
\u6267\u884c\u5b89\u88c5LangChain\u547d\u4ee4\u540e\uff0c\u4f1a\u81ea\u52a8\u5b89\u88c5\u4ee5\u4e0b\u76f8\u5173\u7ec4\u4ef6<\/p>\n\n
Installing collected packages: langsmith, langchain-core, langchain-text-splitters, langchain-community, langchain\n<\/code><\/pre>\n\n
\u66f4\u65b0LangChain<\/p>\n\n
pip install --upgrade langchain\n<\/code><\/pre>\n\n
\u514b\u9686LangChain\u5b58\u50a8\u5e93\u4ece\u6e90\u4ee3\u7801\u5b89\u88c5<\/p>\n\n
pip install -e\n<\/code><\/pre>\n\n
\u5b89\u88c5LangChain\u65f6\u5305\u62ec\u5e38\u7528\u7684\u5f00\u6e90LLM\uff08\u5927\u8bed\u8a00\u6a21\u578b\uff09 \u5e93<\/p>\n\n
pip install langchain[llms]\n<\/code><\/pre>\n\n
\u914d\u7f6e\u73af\u5883\u53d8\u91cf<\/h2>\n
\u5b89\u88c5\u7b2c\u4e09\u65b9\u96c6\u6210\u5e93\uff0c\u4ee5\u4f7f\u7528OpenAI<\/p>\n\n
pip install langchain langchain_openai\n<\/code><\/pre>\n\n
\u8bbe\u7f6eOpenAI\u73af\u5883\u53d8\u91cf<\/p>\n\n
import<\/span> os\n\nos.environ[\"OPENAI_BASE_URL\"<\/span>] = \"https:\/\/xxx.com\/v1\"<\/span>\nos.environ[\"OPENAI_API_KEY\"<\/span>] = \"sk-fDqouTlU62yjkBhF46284543Dc8f42438a9529Df74B4Ce65\"<\/span>\n<\/code><\/pre>\n\n
\u57fa\u672c\u4f7f\u7528<\/h2>\n\n
# \u521d\u59cb\u5316\u6a21\u578b<\/span>\nfrom<\/span> langchain_openai import<\/span> ChatOpenAI\n\nllm = ChatOpenAI()\n\n# \u5b89\u88c5\u5e76\u521d\u59cb\u5316\u9009\u62e9\u7684LLM\uff0c\u5c31\u53ef\u4ee5\u5c1d\u8bd5\u4f7f\u7528\u5b83<\/span>\nllm.invoke(\" LangSmith \u662f\u4ec0\u4e48?\"<\/span>)\n<\/code><\/pre>\n\n\n
AIMessage(content='LangSmith\u662f\u4e00\u4e2a\u865a\u6784\u7684\u540d\u5b57\uff0c\u6ca1\u6709\u5177\u4f53\u7684\u5b9a\u4e49\u6216\u542b\u4e49\u3002\u5b83\u53ef\u80fd\u662f\u4e00\u4e2a\u4eba\u7684\u540d\u5b57\u3001\u4e00\u4e2a\u516c\u53f8\u7684\u540d\u79f0\u6216\u8005\u4e00\u79cd\u4ea7\u54c1\u7684\u54c1\u724c\u3002'<\/span>, response_metadata={'token_usage'<\/span>: {'completion_tokens'<\/span>: 44<\/span>, 'prompt_tokens'<\/span>: 14<\/span>, 'total_tokens'<\/span>: 58<\/span>}, 'model_name'<\/span>: 'gpt-3.5-turbo'<\/span>, 'system_fingerprint'<\/span>: None<\/span>, 'finish_reason'<\/span>: 'stop'<\/span>, 'logprobs'<\/span>: None<\/span>}, id<\/span>='run-f39ca76e-06ef-4815-ba7e-4a4924ef8e48-0'<\/span>)\n<\/code><\/pre>\n\n
\u4f7f\u7528\u63d0\u793a\u6a21\u677f<\/h2>\n
\u4f7f\u7528\u63d0\u793a\u6a21\u677f\u6765\u6307\u5bfc\u5176\u54cd\u5e94\u3002 \u63d0\u793a\u6a21\u677f\u5c06\u539f\u59cb\u7528\u6237\u8f93\u5165\u8f6c\u6362\u4e3a\u66f4\u597d\u7684 LLM \u8f93\u5165<\/p>\n\n
# \u521d\u59cb\u5316\u6a21\u578b<\/span>\nfrom<\/span> langchain_openai import<\/span> ChatOpenAI\n\nllm = ChatOpenAI()\n\n# \u521b\u5efa\u63d0\u793a\u6a21\u677f<\/span>\nfrom<\/span> langchain_core.prompts import<\/span> ChatPromptTemplate\nprompt = ChatPromptTemplate.from_messages([\n    (\"system\"<\/span>, \"\u60a8\u662f\u4e16\u754c\u7ea7\u7684\u6280\u672f\u6587\u6863\u7f16\u5199\u8005\u3002\"<\/span>),\n    (\"user\"<\/span>, \"{input}\"<\/span>)\n])\n\n# \u7ec4\u5408\u6210\u4e00\u4e2a\u7b80\u5355\u7684 LLM \u94fe<\/span>\nchain = prompt | llm \n\n# \u4f7f\u7528LLM\u94fe<\/span>\nchain.invoke({\"input\"<\/span>: \"Langsmith \u5982\u4f55\u5e2e\u52a9\u8fdb\u884c\u6d4b\u8bd5?\"<\/span>})\n<\/code><\/pre>\n\n
\u867d\u7136\u5b83\u4f9d\u7136\u4e0d\u77e5\u9053\u7b54\u6848\uff0c\u4f46\u5bf9\u4e8e\u6280\u672f\u4f5c\u8005\u6765\u8bf4\uff0c\u5b83\u4f7f\u7528\u4e86\u66f4\u6070\u5f53\u7684\u8bed\u6c14\u7ed9\u4e88\u56de\u5e94\u3002<\/p>\n\n
AIMessage(content='Langsmith \u53ef\u4ee5\u5e2e\u52a9\u6d4b\u8bd5\u56e2\u961f\u8fdb\u884c\u6d4b\u8bd5\u7684\u6587\u6863\u7f16\u5199\u5de5\u4f5c\u3002\u4f5c\u4e3a\u4e00\u540d\u4e16\u754c\u7ea7\u7684\u6280\u672f\u6587\u6863\u7f16\u5199\u8005\uff0c\u6211\u53ef\u4ee5\u4e3a\u60a8\u63d0\u4f9b\u4ee5\u4e0b\u5e2e\u52a9\uff1a\\n\\n1. \u7f16\u5199\u6d4b\u8bd5\u8ba1\u5212\uff1a\u6839\u636e\u9879\u76ee\u9700\u6c42\uff0c\u6211\u53ef\u4ee5\u5e2e\u52a9\u60a8\u7f16\u5199\u8be6\u7ec6\u7684\u6d4b\u8bd5\u8ba1\u5212\uff0c\u5305\u62ec\u6d4b\u8bd5\u76ee\u6807\u3001\u6d4b\u8bd5\u8303\u56f4\u3001\u6d4b\u8bd5\u8d44\u6e90\u3001\u6d4b\u8bd5\u7b56\u7565\u7b49\u5185\u5bb9\u3002\\n\\n2. \u7f16\u5199\u6d4b\u8bd5\u7528\u4f8b\uff1a\u6211\u53ef\u4ee5\u5e2e\u52a9\u60a8\u7f16\u5199\u5168\u9762\u7684\u6d4b\u8bd5\u7528\u4f8b\uff0c\u8986\u76d6\u5404\u79cd\u529f\u80fd\u3001\u573a\u666f\u548c\u8fb9\u754c\u6761\u4ef6\u3002\u6d4b\u8bd5\u7528\u4f8b\u5c06\u8be6\u7ec6\u63cf\u8ff0\u6bcf\u4e2a\u6d4b\u8bd5\u6b65\u9aa4\u3001\u9884\u671f\u7ed3\u679c\u548c\u5b9e\u9645\u7ed3\u679c\u3002\\n\\n3. \u7f16\u5199\u6d4b\u8bd5\u62a5\u544a\uff1a\u5728\u6d4b\u8bd5\u5b8c\u6210\u540e\uff0c\u6211\u53ef\u4ee5\u5e2e\u52a9\u60a8\u7f16\u5199\u6e05\u6670\u3001\u8be6\u7ec6\u7684\u6d4b\u8bd5\u62a5\u544a\uff0c\u5305\u62ec\u6d4b\u8bd5\u6267\u884c\u60c5\u51b5\u3001\u53d1\u73b0\u7684\u7f3a\u9677\u3001\u6d4b\u8bd5\u603b\u7ed3\u7b49\u5185\u5bb9\u3002\\n\\n4. \u4f18\u5316\u6d4b\u8bd5\u6587\u6863\uff1a\u5982\u679c\u60a8\u5df2\u7ecf\u6709\u6d4b\u8bd5\u6587\u6863\uff0c\u4f46\u5e0c\u671b\u5bf9\u5176\u8fdb\u884c\u4f18\u5316\uff0c\u4f7f\u5176\u66f4\u52a0\u89c4\u8303\u3001\u6613\u8bfb\u3001\u6613\u7406\u89e3\uff0c\u6211\u4e5f\u53ef\u4ee5\u63d0\u4f9b\u5e2e\u52a9\u3002\\n\\n\u65e0\u8bba\u60a8\u9700\u8981\u54ea\u65b9\u9762\u7684\u5e2e\u52a9\uff0c\u6211\u90fd\u53ef\u4ee5\u6839\u636e\u60a8\u7684\u5177\u4f53\u9700\u6c42\u63d0\u4f9b\u5b9a\u5236\u5316\u7684\u670d\u52a1\u3002\u8bf7\u544a\u8bc9\u6211\u60a8\u7684\u5177\u4f53\u8981\u6c42\uff0c\u6211\u5c06\u5c3d\u529b\u6ee1\u8db3\u60a8\u7684\u9700\u6c42\u3002'<\/span>, response_metadata={'token_usage'<\/span>: {'completion_tokens'<\/span>: 355<\/span>, 'prompt_tokens'<\/span>: 39<\/span>, 'total_tokens'<\/span>: 394<\/span>}, 'model_name'<\/span>: 'gpt-3.5-turbo'<\/span>, 'system_fingerprint'<\/span>: None<\/span>, 'finish_reason'<\/span>: 'stop'<\/span>, 'logprobs'<\/span>: None<\/span>}, id<\/span>='run-b2ed1dbe-b57e-4472-ab19-61536a238c8d-0'<\/span>)\n<\/code><\/pre>\n\n
\u4f7f\u7528\u8f93\u51fa\u89e3\u6790\u5668<\/h2>\n
\u6dfb\u52a0\u4e00\u4e2a\u7b80\u5355\u7684\u8f93\u51fa\u89e3\u6790\u5668\uff0c\u5c06\u804a\u5929\u6d88\u606f\u8f6c\u6362\u4e3a\u5b57\u7b26\u4e32\u3002<\/p>\n\n
from<\/span> langchain_openai import<\/span> ChatOpenAI\nfrom<\/span> langchain_core.prompts import<\/span> ChatPromptTemplate\nfrom<\/span> langchain_core.output_parsers import<\/span> StrOutputParser\n\n# \u521d\u59cb\u5316\u6a21\u578b<\/span>\nllm = ChatOpenAI()\n\n# \u521b\u5efa\u63d0\u793a\u6a21\u677f<\/span>\nprompt = ChatPromptTemplate.from_messages([\n    (\"system\"<\/span>, \"\u60a8\u662f\u4e16\u754c\u7ea7\u7684\u6280\u672f\u6587\u6863\u7f16\u5199\u8005\u3002\"<\/span>),\n    (\"user\"<\/span>, \"{input}\"<\/span>)\n])\n\n# \u4f7f\u7528\u8f93\u51fa\u89e3\u6790\u5668<\/span>\noutput_parser = StrOutputParser()\n\n# \u5c06\u5176\u6dfb\u52a0\u5230\u4e0a\u4e00\u4e2a\u94fe\u4e2d<\/span>\nchain = prompt | llm | output_parser\n\n# \u8c03\u7528\u5b83\u5e76\u63d0\u51fa\u540c\u6837\u7684\u95ee\u9898\u3002\u7b54\u6848\u662f\u4e00\u4e2a\u5b57\u7b26\u4e32\uff0c\u800c\u4e0d\u662fChatMessage<\/span>\nchain.invoke({\"input\"<\/span>: \"Langsmith \u5982\u4f55\u5e2e\u52a9\u8fdb\u884c\u6d4b\u8bd5?\"<\/span>})\n<\/code><\/pre>\n\n\n
'Langsmith \u53ef\u4ee5\u5e2e\u52a9\u6d4b\u8bd5\u56e2\u961f\u8fdb\u884c\u6d4b\u8bd5\u7684\u6587\u6863\u7f16\u5199\u5de5\u4f5c\u3002\u4f5c\u4e3a\u4e00\u540d\u4e16\u754c\u7ea7\u7684\u6280\u672f\u6587\u6863\u7f16\u5199\u8005\uff0c\u6211\u53ef\u4ee5\u4e3a\u60a8\u63d0\u4f9b\u4ee5\u4e0b\u5e2e\u52a9\uff1a\\n\\n1. \u7f16\u5199\u6d4b\u8bd5\u8ba1\u5212\uff1a\u6839\u636e\u9879\u76ee\u9700\u6c42\uff0c\u6211\u53ef\u4ee5\u5e2e\u52a9\u60a8\u7f16\u5199\u8be6\u7ec6\u7684\u6d4b\u8bd5\u8ba1\u5212\uff0c\u5305\u62ec\u6d4b\u8bd5\u76ee\u6807\u3001\u6d4b\u8bd5\u8303\u56f4\u3001\u6d4b\u8bd5\u8d44\u6e90\u3001\u6d4b\u8bd5\u7b56\u7565\u7b49\u5185\u5bb9\u3002\\n\\n2. \u7f16\u5199\u6d4b\u8bd5\u7528\u4f8b\uff1a\u6211\u53ef\u4ee5\u5e2e\u52a9\u60a8\u7f16\u5199\u5168\u9762\u7684\u6d4b\u8bd5\u7528\u4f8b\uff0c\u8986\u76d6\u5404\u79cd\u529f\u80fd\u3001\u573a\u666f\u548c\u8fb9\u754c\u6761\u4ef6\u3002\u6d4b\u8bd5\u7528\u4f8b\u5c06\u8be6\u7ec6\u63cf\u8ff0\u6bcf\u4e2a\u6d4b\u8bd5\u6b65\u9aa4\u3001\u9884\u671f\u7ed3\u679c\u548c\u5b9e\u9645\u7ed3\u679c\u3002\\n\\n3. \u7f16\u5199\u6d4b\u8bd5\u62a5\u544a\uff1a\u5728\u6d4b\u8bd5\u5b8c\u6210\u540e\uff0c\u6211\u53ef\u4ee5\u5e2e\u52a9\u60a8\u7f16\u5199\u6e05\u6670\u3001\u8be6\u7ec6\u7684\u6d4b\u8bd5\u62a5\u544a\uff0c\u5305\u62ec\u6d4b\u8bd5\u6267\u884c\u60c5\u51b5\u3001\u53d1\u73b0\u7684\u7f3a\u9677\u3001\u6d4b\u8bd5\u603b\u7ed3\u7b49\u5185\u5bb9\u3002\\n\\n4. \u4f18\u5316\u6d4b\u8bd5\u6587\u6863\uff1a\u5982\u679c\u60a8\u5df2\u7ecf\u6709\u6d4b\u8bd5\u6587\u6863\uff0c\u4f46\u5e0c\u671b\u5bf9\u5176\u8fdb\u884c\u4f18\u5316\uff0c\u4f7f\u5176\u66f4\u52a0\u89c4\u8303\u3001\u6613\u8bfb\u3001\u6613\u7406\u89e3\uff0c\u6211\u4e5f\u53ef\u4ee5\u63d0\u4f9b\u5e2e\u52a9\u3002\\n\\n\u65e0\u8bba\u60a8\u9700\u8981\u54ea\u65b9\u9762\u7684\u5e2e\u52a9\uff0c\u6211\u90fd\u53ef\u4ee5\u6839\u636e\u60a8\u7684\u5177\u4f53\u9700\u6c42\u63d0\u4f9b\u5b9a\u5236\u5316\u7684\u670d\u52a1\u3002\u8bf7\u544a\u8bc9\u6211\u60a8\u7684\u5177\u4f53\u8981\u6c42\uff0c\u6211\u5c06\u5c3d\u529b\u6ee1\u8db3\u60a8\u7684\u9700\u6c42\u3002'<\/span>\n<\/code><\/pre>\n\n
\u5411\u91cf\u5b58\u50a8<\/h2>\n
\u52a0\u8f7d\u8981\u7d22\u5f15\u7684\u6570\u636e,\u9700\u8981\u5b89\u88c5BeautifulSoup<\/p>\n\n
pip install beautifulsoup4\n<\/code><\/pre>\n\n
\u5c06\u5176\u7d22\u5f15\u5230\u5411\u91cf\u5b58\u50a8\u4e2d\u3002\u8fd9\u9700\u8981\u4e00\u4e9b\u7ec4\u4ef6\uff0c\u5373\u5d4c\u5165\u6a21\u578b\u548c\u5411\u91cf\u5b58\u50a8\u3002<\/p>\n
\n
\u4f7f\u7528\u4e00\u4e2a\u7b80\u5355\u7684\u672c\u5730\u5411\u91cf\u5b58\u50a8 FAISS\uff0c\u9996\u5148\u9700\u8981\u5b89\u88c5\u5b83<\/p>\n<\/blockquote>\n\n
pip install faiss-cpu\n<\/code><\/pre>\n\n\n
# \u5bfc\u5165\u548c\u4f7f\u7528 WebBaseLoader<\/span>\nfrom<\/span> langchain_community.document_loaders import<\/span> WebBaseLoader\n\nloader = WebBaseLoader(\"https:\/\/docs.smith.langchain.com\/user_guide\"<\/span>)\ndocs = loader.load()\n\n# \u5bf9\u4e8e\u5d4c\u5165\u6a21\u578b\uff0c\u8fd9\u91cc\u901a\u8fc7 API\u8c03\u7528<\/span>\nfrom<\/span> langchain_openai import<\/span> OpenAIEmbeddings\n\nembeddings = OpenAIEmbeddings()\n\n#\u4f7f\u7528\u6b64\u5d4c\u5165\u6a21\u578b\u5c06\u6587\u6863\u6444\u53d6\u5230\u77e2\u91cf\u5b58\u50a8\u4e2d<\/span>\nfrom<\/span> langchain_community.vectorstores import<\/span> FAISS\nfrom<\/span> langchain_text_splitters import<\/span> RecursiveCharacterTextSplitter\n\n# \u4f7f\u7528\u5206\u5272\u5668\u5206\u5272\u6587\u6863<\/span>\ntext_splitter = RecursiveCharacterTextSplitter()\ndocuments = text_splitter.split_documents(docs)\n# \u5411\u91cf\u5b58\u50a8<\/span>\nvector = FAISS.from_documents(documents, embeddings)\n<\/code><\/pre>\n\n
\u68c0\u7d22\u94fe<\/h2>\n
\n
\u5df2\u5728\u5411\u91cf\u5b58\u50a8\u4e2d\u7d22\u5f15\u4e86\u8fd9\u4e9b\u6570\u636e\uff0c\u63a5\u4e0b\u6765\u8981\u521b\u5efa\u4e00\u4e2a\u68c0\u7d22\u94fe\u3002\u8be5\u94fe\u5c06\u63a5\u6536\u4e00\u4e2a\u4f20\u5165\u7684\u95ee\u9898\uff0c\u67e5\u627e\u76f8\u5173\u6587\u6863\uff0c\u7136\u540e\u5c06\u8fd9\u4e9b\u6587\u6863\u4e0e\u539f\u59cb\u95ee\u9898\u4e00\u8d77\u4f20\u9012\u7ed9LLM\uff0c\u8981\u6c42\u5b83\u56de\u7b54\u539f\u59cb\u95ee\u9898\u3002<\/p>\n<\/blockquote>\n
\u521b\u5efa\u4e00\u4e2a\u94fe\uff0c\u8be5\u94fe\u63a5\u53d7\u4e00\u4e2a\u95ee\u9898\u548c\u68c0\u7d22\u5230\u7684\u6587\u6863\u5e76\u751f\u6210\u4e00\u4e2a\u7b54\u6848\u3002<\/p>\n\n
from<\/span> langchain.chains.combine_documents import<\/span> create_stuff_documents_chain\n\nprompt = ChatPromptTemplate.from_template(\"\"\"\u4ec5\u6839\u636e\u63d0\u4f9b\u7684\u4e0a\u4e0b\u6587\u56de\u7b54\u4ee5\u4e0b\u95ee\u9898:\n\n\n{context}\n<\/context>\n\nQuestion: {input}\"\"\"<\/span>)\n\n# \u521b\u5efa\u94fe\uff0c\u8be5\u94fe\u83b7\u53d6\u6587\u6863\u5217\u8868\u5e76\u5c06\u5b83\u4eec\u5168\u90e8\u683c\u5f0f\u5316\u4e3a\u63d0\u793a\uff0c\u7136\u540e\u5c06\u8be5\u63d0\u793a\u4f20\u9012\u7ed9LLM\u3002\u5b83\u4f20\u9012\u6240\u6709\u6587\u6863\uff0c\u56e0\u6b64\u5e94\u8be5\u786e\u4fdd\u5b83\u9002\u5408\u6b63\u5728\u4f7f\u7528\u7684 LLM \u4e0a\u4e0b\u6587\u7a97\u53e3<\/span>\ndocument_chain = create_stuff_documents_chain(llm, prompt)\n\n# \u53ef\u4ee5\u76f4\u63a5\u901a\u8fc7\u4f20\u5165\u6587\u6863\u6765\u8fd0\u884c\u5b83<\/span>\nfrom<\/span> langchain_core.documents import<\/span> Document\n\ntext =\"langsmith can let you visualize test results\"<\/span>\ndocument_chain.invoke({\n    \"input\"<\/span>: \"Langsmith \u5982\u4f55\u5e2e\u52a9\u8fdb\u884c\u6d4b\u8bd5?\"<\/span>,\n    \"context\"<\/span>: [Document(page_content=text)]\n})\n<\/code><\/pre>\n\n\n
'LangSmith \u63d0\u4f9b\u4e86\u591a\u79cd\u65b9\u5f0f\u6765\u5e2e\u52a9\u8fdb\u884c\u6d4b\u8bd5\u3002\\n\\n\u9996\u5148\uff0cLangSmith \u652f\u6301\u5f00\u53d1\u4eba\u5458\u521b\u5efa\u6570\u636e\u96c6\uff0c\u8fd9\u4e9b\u6570\u636e\u96c6\u662f\u8f93\u5165\u548c\u53c2\u8003\u8f93\u51fa\u7684\u96c6\u5408\uff0c\u5e76\u4f7f\u7528\u8fd9\u4e9b\u6570\u636e\u96c6\u5728\u4ed6\u4eec\u7684 LLM \u5e94\u7528\u7a0b\u5e8f\u4e0a\u8fd0\u884c\u6d4b\u8bd5\u3002\u5f00\u53d1\u4eba\u5458\u53ef\u4ee5\u6279\u91cf\u4e0a\u4f20\u3001\u52a8\u6001\u521b\u5efa\u6216\u4ece\u5e94\u7528\u7a0b\u5e8f\u8ddf\u8e2a\u4e2d\u5bfc\u51fa\u6d4b\u8bd5\u7528\u4f8b\u3002\u6b64\u5916\uff0cLangSmith \u8fd8\u53ef\u4ee5\u8f7b\u677e\u8fd0\u884c\u81ea\u5b9a\u4e49\u8bc4\u4f30\u6765\u5bf9\u6d4b\u8bd5\u7ed3\u679c\u8fdb\u884c\u8bc4\u5206\u3002\\n\\n\u5176\u6b21\uff0cLangSmith \u63d0\u4f9b\u6bd4\u8f83\u89c6\u56fe\uff0c\u53ef\u4ee5\u5e76\u6392\u67e5\u770b\u540c\u4e00\u6570\u636e\u70b9\u4e0a\u4e0d\u540c\u914d\u7f6e\u7684\u7ed3\u679c\u3002\u8fd9\u5bf9\u4e8e\u5bf9\u5e94\u7528\u7a0b\u5e8f\u7684\u4e0d\u540c\u7248\u672c\u8fdb\u884c\u539f\u578b\u8bbe\u8ba1\u548c\u66f4\u6539\u65f6\u975e\u5e38\u6709\u7528\uff0c\u53ef\u4ee5\u5e2e\u52a9\u5f00\u53d1\u4eba\u5458\u4e86\u89e3\u54ea\u4e2a\u53d8\u4f53\u7684\u6027\u80fd\u66f4\u597d\u3002\\n\\n\u6b64\u5916\uff0cLangSmith \u8fd8\u63d0\u4f9b\u4e86\u4e00\u4e2a Playground \u73af\u5883\uff0c\u53ef\u4ee5\u7528\u4e8e\u5feb\u901f\u8fed\u4ee3\u548c\u5b9e\u9a8c\u3002\u5f00\u53d1\u4eba\u5458\u53ef\u4ee5\u5728 Playground \u4e2d\u5feb\u901f\u6d4b\u8bd5\u4e0d\u540c\u7684\u63d0\u793a\u548c\u6a21\u578b\uff0c\u5e76\u5c06\u6bcf\u6b21\u8fd0\u884c\u8bb0\u5f55\u5728\u7cfb\u7edf\u4e2d\u4ee5\u4f9b\u540e\u7eed\u521b\u5efa\u6d4b\u8bd5\u7528\u4f8b\u6216\u4e0e\u5176\u4ed6\u8fd0\u884c\u8fdb\u884c\u6bd4\u8f83\u3002\\n\\n\u6700\u540e\uff0cLangSmith \u8fd8\u652f\u6301\u81ea\u52a8\u5316\uff0c\u53ef\u4ee5\u8fd1\u4e4e\u5b9e\u65f6\u5730\u5bf9\u8ddf\u8e2a\u6267\u884c\u64cd\u4f5c\u3002\u5f00\u53d1\u4eba\u5458\u53ef\u4ee5\u5b9a\u4e49\u81ea\u52a8\u5316\u64cd\u4f5c\uff0c\u5305\u62ec\u8bc4\u5206\u3001\u53d1\u9001\u5230\u6ce8\u91ca\u961f\u5217\u6216\u6dfb\u52a0\u5230\u6570\u636e\u96c6\u7b49\u3002\u8fd9\u5bf9\u4e8e\u5728\u751f\u4ea7\u89c4\u6a21\u4e0a\u5904\u7406\u8ddf\u8e2a\u975e\u5e38\u6709\u7528\u3002'<\/span>\n<\/code><\/pre>\n\n
\u8fd8\u53ef\u4ee5\u8ba9\u6587\u6863\u9996\u5148\u6765\u81ea\u521a\u521a\u8bbe\u7f6e\u7684\u68c0\u7d22\u5668\u3002 \u8fd9\u6837\uff0c\u53ef\u4ee5\u4f7f\u7528\u68c0\u7d22\u5668\u52a8\u6001\u9009\u62e9\u6700\u76f8\u5173\u7684\u6587\u6863\uff0c\u5e76\u5c06\u5176\u4f20\u9012\u7ed9\u7ed9\u5b9a\u7684\u95ee\u9898\u3002<\/p>\n\n
from<\/span> langchain.chains import<\/span> create_retrieval_chain\n\n# \u521b\u5efa\u5411\u91cf\u5b58\u50a8\u68c0\u7d22\u5668<\/span>\nretriever = vector.as_retriever()\n# \u521b\u5efa\u94fe\uff0c\u8be5\u94fe\u63a5\u6536\u7528\u6237\u67e5\u8be2\uff0c\u7136\u540e\u5c06\u5176\u4f20\u9012\u7ed9\u68c0\u7d22\u5668\u4ee5\u83b7\u53d6\u76f8\u5173\u6587\u6863\u3002\u7136\u540e\u5c06\u8fd9\u4e9b\u6587\u6863\uff08\u548c\u539f\u59cb\u8f93\u5165\uff09\u4f20\u9012\u5230 LLM \u4ee5\u751f\u6210\u54cd\u5e94<\/span>\nretrieval_chain = create_retrieval_chain(retriever, document_chain)\n# \u6267\u884c\u68c0\u7d22 \u8fd9\u5c06\u8fd4\u56de\u4e00\u4e2a\u5b57\u5178<\/span>\nresponse = retrieval_chain.invoke({\"input\"<\/span>: \"how can langsmith help with testing?\"<\/span>})\nprint<\/span>(response[\"answer\"<\/span>])\n<\/code><\/pre>\n\n
\u7b54\u6848\u5e94\u8be5\u66f4\u51c6\u786e<\/p>\n\n
LangSmith can help<\/span> with<\/span> testing in<\/span> several ways. \n\n1.<\/span> LangSmith allows developers to create datasets, which are collections of inputs and<\/span> reference outputs, and<\/span> use these to run tests on their LLM applications. Test cases can be uploaded in<\/span> bulk, created on the fly, or<\/span> exported from<\/span> application traces.\n\n2.<\/span> LangSmith provides a user-friendly comparison view for<\/span> test runs. This allows developers to compare the results of different configurations on the same datapoints side-by-side, helping them identify any<\/span> regressions or<\/span> improvements.\n\n3.<\/span> LangSmith supports custom evaluations, both LLM-based and<\/span> heuristic-based, to score test results.\n\nOverall, LangSmith enables developers to perform test-driven development and<\/span> evaluate the performance of their LLM applications during the prototyping and<\/span> beta testing phases.\n<\/code><\/pre>\n\n
\u5bf9\u8bdd\u68c0\u7d22\u94fe<\/h2>\n
\u4e0a\u9762\u521b\u5efa\u7684\u94fe\u53ea\u80fd\u56de\u7b54\u5355\u4e2a\u95ee\u9898\u3002\u73b0\u5728\u521b\u5efa\u4e00\u4e2a\u65b0\u94fe\u3002\u8be5\u94fe\u5c06\u63a5\u6536\u6700\u65b0\u7684\u8f93\u5165\u548c\u5bf9\u8bdd\u5386\u53f2\u8bb0\u5f55\uff0c\u5e76\u4f7f\u7528 LLM \u751f\u6210\u641c\u7d22\u67e5\u8be2\u3002<\/p>\n\n
from<\/span> langchain.chains import<\/span> create_history_aware_retriever\nfrom<\/span> langchain_core.prompts import<\/span> MessagesPlaceholder\n\n# First we need a prompt that we can pass into an LLM to generate this search query<\/span>\n\nprompt = ChatPromptTemplate.from_messages([\n    MessagesPlaceholder(variable_name=\"chat_history\"<\/span>),\n    (\"user\"<\/span>, \"{input}\"<\/span>),\n    (\"user\"<\/span>, \"\u9274\u4e8e\u4e0a\u8ff0\u5bf9\u8bdd\uff0c\u751f\u6210\u4e00\u4e2a\u641c\u7d22\u67e5\u8be2\u4ee5\u67e5\u627e\u4ee5\u83b7\u53d6\u4e0e\u5bf9\u8bdd\u76f8\u5173\u7684\u4fe1\u606f\"<\/span>)\n])\nretriever_chain = create_history_aware_retriever(llm, retriever, prompt)\n\n# \u901a\u8fc7\u4f20\u5165\u7528\u6237\u63d0\u51fa\u540e\u7eed\u95ee\u9898\u6765\u6d4b\u8bd5<\/span>\nfrom<\/span> langchain_core.messages import<\/span> HumanMessage, AIMessage\n\nchat_history = [HumanMessage(content=\"LangSmith \u53ef\u4ee5\u5e2e\u52a9\u6d4b\u8bd5\u6211\u7684 LLM \u5e94\u7528\u7a0b\u5e8f\u5417?\"<\/span>), AIMessage(content=\"Yes!\"<\/span>)]\nretriever_chain.invoke({\n    \"chat_history\"<\/span>: chat_history,\n    \"input\"<\/span>: \"\u544a\u8bc9\u6211\u600e\u4e48\u505a\"<\/span>\n})\n<\/code><\/pre>\n\n
\u8fd8\u53ef\u4ee5\u521b\u5efa\u4e00\u4e2a\u65b0\u7684\u94fe\u6765\u7ee7\u7eed\u5bf9\u8bdd\uff0c\u5e76\u7262\u8bb0\u8fd9\u4e9b\u68c0\u7d22\u5230\u7684\u6587\u6863\u3002<\/p>\n\n
prompt = ChatPromptTemplate.from_messages([\n    (\"system\"<\/span>, \"\u6839\u636e\u4ee5\u4e0b\u4e0a\u4e0b\u6587\u56de\u7b54\u7528\u6237\u7684\u95ee\u9898:\\n\\n{context}\"<\/span>),\n    MessagesPlaceholder(variable_name=\"chat_history\"<\/span>),\n    (\"user\"<\/span>, \"{input}\"<\/span>),\n])\n\ndocument_chain = create_stuff_documents_chain(llm, prompt)\nretrieval_chain = create_retrieval_chain(retriever_chain, document_chain)\n\n# \u6d4b\u8bd5<\/span>\nchat_history = [HumanMessage(content=\"LangSmith \u53ef\u4ee5\u5e2e\u52a9\u6d4b\u8bd5\u6211\u7684 LLM \u5e94\u7528\u7a0b\u5e8f\u5417?\"<\/span>), AIMessage(content=\"Yes!\"<\/span>)]\nretrieval_chain.invoke({\n    \"chat_history\"<\/span>: chat_history,\n    \"input\"<\/span>: \"Tell me how\"<\/span>\n})\n<\/code><\/pre>\n\n
\u4ee3\u7406\u7684\u4f7f\u7528<\/h1>\n
\u6784\u5efa\u4ee3\u7406\u65f6\u8981\u505a\u7684\u7b2c\u4e00\u4ef6\u4e8b\u662f\u786e\u5b9a\u5b83\u5e94\u8be5\u6709\u6743\u8bbf\u95ee\u54ea\u4e9b\u5de5\u5177\u3002\u8fd9\u91cc\u6388\u4e88\u4ee3\u7406\u8bbf\u95ee\u4e24\u4e2a\u5de5\u5177\u7684\u6743\u9650\uff1a<\/p>\n\n
\u4f7f\u7528\u521b\u5efa\u7684\u68c0\u7d22\u5668\uff0c\u4ee5\u4fbf\u4ee3\u7406\u80fd\u591f\u56de\u7b54\u6709\u5173LangSmith\u7684\u95ee\u9898\u3002\n\n\u4e00\u4e2a\u641c\u7d22\u5de5\u5177\uff0c\u4ee5\u4f7f\u4ee3\u7406\u80fd\u591f\u56de\u7b54\u9700\u8981\u6700\u65b0\u4fe1\u606f\u7684\u95ee\u9898\u3002\n<\/code><\/pre>\n\n
\u68c0\u7d22\u5668\u5de5\u5177<\/h2>\n\n
from<\/span> langchain.tools.retriever import<\/span> create_retriever_tool\n\nretriever = vector.as_retriever()\nretriever_tool = create_retriever_tool(\n    retriever,\n    \"langsmith_search\"<\/span>,\n    \"\u641c\u7d22\u6709\u5173 LangSmith \u7684\u4fe1\u606f\u3002\u5bf9\u4e8e\u6709\u5173LangSmith\u7684\u4efb\u4f55\u95ee\u9898\uff0c\u60a8\u5fc5\u987b\u4f7f\u7528\u6b64\u5de5\u5177!\"<\/span>,\n)\n<\/code><\/pre>\n\n
\u521b\u5efa\u641c\u7d22\u5de5\u5177<\/h2>\n
\u8bbf\u95eeTavily<\/a>\uff0c\u6ce8\u518c\u8d26\u53f7\u767b\u5f55\u5e76\u521b\u5efaAPI\u79d8\u94a5\uff0c\u7136\u540e\u914d\u7f6e\u73af\u5883\u53d8\u91cf<\/p>\n\n
import<\/span> os\n\nos.environ[\"TAVILY_API_KEY\"<\/span>] = 'tvly-ScxxxxxxxM8'<\/span>\n<\/code><\/pre>\n\n
\u5b89\u88c5tavily-python\u5e93<\/p>\n\n
pip install -U langchain-community tavily-python\n<\/code><\/pre>\n\n
\u521b\u5efa\u5de5\u5177<\/p>\n\n
from<\/span> langchain_community.tools.tavily_search import<\/span> TavilySearchResults\n\nsearch = TavilySearchResults()\n<\/code><\/pre>\n\n
\u521b\u5efa\u4ee3\u7406<\/h2>\n
\u521b\u5efa\u4f7f\u7528\u5de5\u5177\u7684\u5217\u8868<\/p>\n\n
tools = [retriever_tool, search]\n<\/code><\/pre>\n\n
\u521b\u5efa\u4e00\u4e2a\u4ee3\u7406\u6765\u4f7f\u7528\u5de5\u5177<\/p>\n\n
from<\/span> langchain_openai import<\/span> ChatOpenAI\nfrom<\/span> langchain import<\/span> hub\nfrom<\/span> langchain.agents import<\/span> create_openai_functions_agent\nfrom<\/span> langchain.agents import<\/span> AgentExecutor\n\n# \u83b7\u53d6\u4f7f\u7528\u63d0\u793a \u53ef\u4ee5\u4fee\u6539\u5b83<\/span>\nprompt = hub.pull(\"hwchase17\/openai-functions-agent\"<\/span>)\n# \u521d\u59cb\u5316\u5927\u6a21\u578b<\/span>\nllm = ChatOpenAI(model=\"gpt-3.5-turbo\"<\/span>, temperature=0<\/span>)\n# \u521b\u5efa\u4e00\u4e2aopenai_functions_agent\u4ee3\u7406<\/span>\nagent = create_openai_functions_agent(llm, tools, prompt)\n# \u521b\u5efa\u4ee3\u7406\u6267\u884c\u5668<\/span>\nagent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True<\/span>)\n<\/code><\/pre>\n\n
\u4f7f\u7528Agent\u4ee3\u7406<\/h2>\n\n
# \u6267\u884cAgent<\/span>\nagent_executor.invoke({\"input\"<\/span>: \"Langsmith \u5982\u4f55\u5e2e\u52a9\u8fdb\u884c\u6d4b\u8bd5?\"<\/span>})\n<\/code><\/pre>\n\n
\n
\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/figure>\n\n\u8be2\u95ee\u5929\u6c14\u60c5\u51b5<\/p>\n\n
# \u6267\u884cAgent<\/span>\nagent_executor.invoke({\"input\"<\/span>: \"\u6210\u90fd\u4eca\u5929\u5929\u6c14\u60c5\u51b5?\"<\/span>})\n<\/code><\/pre>\n\n\n
> Entering new AgentExecutor chain...\n\nInvoking: `tavily_search_results_json` with<\/span> `{'query'<\/span>: '\u6210\u90fd\u4eca\u5929\u5929\u6c14\u60c5\u51b5'<\/span>}`\n\n[{'url'<\/span>: 'http:\/\/www.nmc.cn\/publish\/forecast\/ASC\/chengdu.html'<\/span>, 'content'<\/span>: '\u6210\u90fd\u5929\u6c14\u9884\u62a5 ; \u7701\u4efd\uff1a \u57ce\u5e02\uff1a ... \u5236\u4f5c\u7ef4\u62a4\uff1a\u56fd\u5bb6\u6c14\u8c61\u4e2d\u5fc3\u9884\u62a5\u7cfb\u7edf\u5f00\u653e\u5b9e\u9a8c\u5ba4 \u5730\u5740\uff1a\u5317\u4eac\u5e02\u4e2d\u5173\u6751\u5357\u5927\u885746\u53f7 \u90ae\u7f16\uff1a100081 . \u4eac\u516c\u7f51\u5b89\u5907 11040102700100 ...'<\/span>},\n{'url'<\/span>: 'http:\/\/www.weather.com.cn\/weather\/101270101.shtml'<\/span>, 'content'<\/span>: '\u6d82\u64e6SPF\u5927\u4e8e15\u3001PA+\u9632\u6652\u62a4\u80a4\u54c1\u3002\\n\u5929\u51c9\uff0c\u6e7f\u5ea6\u5927\uff0c\u8f83\u6613\u611f\u5192\u3002\\n\u5929\u6c14\u51c9\uff0c\u5728\u6237\u5916\u8fd0\u52a8\u8bf7\u6ce8\u610f\u589e\u51cf\u8863\u7269\u3002\\n\u65e0\u9700\u62c5\u5fc3\u8fc7\u654f\uff0c\u53ef\u653e\u5fc3\u5916\u51fa\uff0c\u4eab\u53d7\u751f\u6d3b\u3002\\n\u5efa\u8bae\u7740\u539a\u5916\u5957\u52a0\u6bdb\u8863\u7b49\u670d\u88c5\u3002\\n\u5929\u6c14\u8f83\u597d\uff0c\u9002\u5408\u64e6\u6d17\u6c7d\u8f66\u3002\\n\u8f90\u5c04\u5f31\uff0c\u6d82\u64e6\n\n\u6839\u636e\u5929\u6c14\u9884\u62a5\uff0c\u6210\u90fd\u4eca\u5929\u7684\u5929\u6c14\u60c5\u51b5\u4e3a\u6674\uff0c\u6c14\u6e29\u4e3a0\u2103\uff0c\u98ce\u529b\u5c0f\u4e8e3\u7ea7\u3002\u660e\u5929\u5c06\u8f6c\u4e3a\u591a\u4e91\uff0c\u6700\u9ad8\u6c14\u6e2912\u2103\uff0c\u6700\u4f4e\u6c14\u6e290\u2103\uff0c\u98ce\u529b\u5c0f\u4e8e3\u7ea7\u3002\n\n> Finished chain.\n<\/span><\/code><\/pre>\n\n
\u8fdb\u884c\u5bf9\u8bdd<\/p>\n\n
from<\/span> langchain_core.messages import<\/span> SystemMessage, HumanMessage, AIMessage\n\nchat_history = [HumanMessage(content=\"LangSmith \u53ef\u4ee5\u5e2e\u52a9\u6d4b\u8bd5\u6211\u7684 LLM \u5e94\u7528\u7a0b\u5e8f\u5417?\"<\/span>), AIMessage(content=\"Yes!\"<\/span>)]\nagent_executor.invoke({\n    \"chat_history\"<\/span>: chat_history,\n    \"input\"<\/span>: \"\u544a\u8bc9\u6211\u600e\u4e48\u505a\"<\/span>\n})\n<\/code><\/pre>\n\n
LangServe\u63d0\u4f9b\u670d\u52a1<\/h1>\n
\u6982\u8ff0<\/h2>\n
\n
LangServe\u53ef\u4ee5\u5e2e\u52a9\u5f00\u53d1\u4eba\u5458\u5c06LangChain\u5e94\u7528\u7a0b\u5e8f\u90e8\u7f72\u4e3aREST API\u3002\u4f7f\u7528LangChain\u65f6\u4e0d\u662f\u5fc5\u5b9a\u4f7f\u7528LangServe\u3002<\/p>\n<\/blockquote>\n
\u5b89\u88c5langserve<\/p>\n\n
pip install \"langserve[all]\"<\/span>\n<\/code><\/pre>\n\n
\u521b\u5efa\u670d\u52a1<\/h2>\n
\u521b\u5efa\u4e00\u4e2aserve.py\u6587\u4ef6\u3002\u5305\u542b\u4e3a\u5e94\u7528\u7a0b\u5e8f\u63d0\u4f9b\u670d\u52a1\u7684\u903b\u8f91\u3002\u7531\u4e09\u90e8\u5206\u7ec4\u6210\uff1a<\/p>\n\n
\u6784\u5efa\u7684\u94fe\u7684\u5b9a\u4e49\n\nFastAPI\u5e94\u7528\u7a0b\u5e8f\n\n\u4e3a\u94fe\u63d0\u4f9b\u670d\u52a1\u7684\u8def\u7531\u7684\u5b9a\u4e49\uff0c\u7531langserve.add_routes\u547d\u4ee4\u5b8c\u6210\n<\/code><\/pre>\n\n\n
import<\/span> os\nfrom<\/span> typing import<\/span> List<\/span>\n\nfrom<\/span> fastapi import<\/span> FastAPI\nfrom<\/span> langchain import<\/span> hub\nfrom<\/span> langchain.agents import<\/span> AgentExecutor\nfrom<\/span> langchain.agents import<\/span> create_openai_functions_agent\nfrom<\/span> langchain.pydantic_v1 import<\/span> BaseModel, Field\nfrom<\/span> langchain.tools.retriever import<\/span> create_retriever_tool\nfrom<\/span> langchain_community.document_loaders import<\/span> WebBaseLoader\nfrom<\/span> langchain_community.tools.tavily_search import<\/span> TavilySearchResults\nfrom<\/span> langchain_community.vectorstores import<\/span> FAISS\nfrom<\/span> langchain_core.messages import<\/span> BaseMessage\nfrom<\/span> langchain_openai import<\/span> ChatOpenAI\nfrom<\/span> langchain_openai import<\/span> OpenAIEmbeddings\nfrom<\/span> langchain_text_splitters import<\/span> RecursiveCharacterTextSplitter\nfrom<\/span> langserve import<\/span> add_routes\n\nos.environ[\"TAVILY_API_KEY\"<\/span>] = 'tvly-Scx77MxxxxIM8'<\/span>\nos.environ[\"OPENAI_BASE_URL\"<\/span>] = \"https:\/\/xxx.com\/v1\"<\/span>\nos.environ[\"OPENAI_API_KEY\"<\/span>] = \"sk-BGFnOL9Q4c99B378B66cT3BlBKFJ28839b4813bc437B82c2\"<\/span>\n\n# 1. \u83b7\u53d6\u68c0\u7d22\u5668<\/span>\n# \u521b\u5efa\u4e00\u4e2a WebBaseLoader \u5bf9\u8c61\uff0c\u52a0\u8f7d\u7ed9\u5b9a URL \u7684\u7f51\u9875\u5185\u5bb9<\/span>\nloader = WebBaseLoader(\"https:\/\/docs.smith.langchain.com\/user_guide\"<\/span>)\n# \u8f7d\u5165\u7f51\u9875\u5185\u5bb9<\/span>\ndocs = loader.load()\n\n# \u521d\u59cb\u5316 RecursiveCharacterTextSplitter \u5bf9\u8c61\u7528\u4e8e\u6587\u672c\u62c6\u5206<\/span>\ntext_splitter = RecursiveCharacterTextSplitter()\n# \u4f7f\u7528\u6587\u672c\u62c6\u5206\u5668\u5c06\u6587\u6863\u5206\u6210\u6bb5\u843d<\/span>\ndocuments = text_splitter.split_documents(docs)\n\n# \u521d\u59cb\u5316 OpenAIEmbeddings \u5bf9\u8c61\uff0c\u7528\u4e8e\u83b7\u53d6\u6587\u672c\u5d4c\u5165<\/span>\nembeddings = OpenAIEmbeddings()\n# \u4ece\u6587\u6863\u4e2d\u83b7\u53d6\u5d4c\u5165\u5411\u91cf\u5e76\u5b58\u50a8<\/span>\nvector = FAISS.from_documents(documents, embeddings)\n\n# \u5c06\u5411\u91cf\u5bf9\u8c61\u8f6c\u6362\u4e3a\u68c0\u7d22\u5668<\/span>\nretriever = vector.as_retriever()\n\n# 2. \u521b\u5efa\u5de5\u5177<\/span>\n# \u68c0\u7d22\u5668\u5de5\u5177<\/span>\nretriever_tool = create_retriever_tool(\n    retriever,\n    \"langsmith_search\"<\/span>,\n    \"Search for information about LangSmith. For any questions about LangSmith, you must use this tool!\"<\/span>,\n)\n\n# \u641c\u7d22\u5de5\u5177<\/span>\nsearch = TavilySearchResults()\ntools = [retriever_tool, search]\n\n# 3.\u521b\u5efa\u4ee3\u7406<\/span>\n# \u4ece\u6307\u5b9a\u7684 Hub \u62c9\u53d6\u63d0\u793a\u6a21\u677f<\/span>\nprompt = hub.pull(\"hwchase17\/openai-functions-agent\"<\/span>)\n\n# \u521d\u59cb\u5316 ChatOpenAI \u5bf9\u8c61\uff0c\u9009\u62e9\u6a21\u578b\u4e3a\"gpt-3.5-turbo\"\uff0c\u8bbe\u7f6e\u6e29\u5ea6\u4e3a0<\/span>\nllm = ChatOpenAI(model=\"gpt-3.5-turbo\"<\/span>, temperature=0<\/span>)\n\n# \u4f7f\u7528\u63d0\u4f9b\u7684\u6a21\u578b\u3001\u5de5\u5177\u548c\u63d0\u793a\u521b\u5efa OpenAI \u51fd\u6570\u4ee3\u7406\u5668<\/span>\nagent = create_openai_functions_agent(llm, tools, prompt)\n\n# \u521d\u59cb\u5316 AgentExecutor\uff0c\u4f20\u5165\u4ee3\u7406\u5668\u3001\u5de5\u5177\u5bf9\u8c61\u548c verbose \u6807\u8bb0\u4e3a True<\/span>\nagent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True<\/span>)\n\n# 4. \u5e94\u7528\u5b9a\u4e49<\/span>\napp = FastAPI(\n    title=\"LangChain Server\"<\/span>,\n    version=\"1.0\"<\/span>,\n    description=\"A simple API server using LangChain's Runnable interfaces\"<\/span>,\n)\n\n\n# 5. \u6dfb\u52a0\u8def\u7531<\/span>\nclass<\/span> Input<\/span>(BaseModel<\/span>):\n    # \u5b9a\u4e49\u8f93\u5165 BaseModel \u5305\u542b\u5b57\u6bb5 input \u548c chat_history<\/span>\n    input<\/span>: str<\/span>\n    chat_history: List<\/span>[BaseMessage] = Field(\n        ...,\n        # \u4e3a chat_history \u5b57\u6bb5\u6dfb\u52a0\u989d\u5916\u5c5e\u6027\uff0c\u8bbe\u7f6e type \u4e3a \"chat\"\uff0cinput \u4e3a \"location\"<\/span>\n        extra={\"widget\"<\/span>: {\"type\"<\/span>: \"chat\"<\/span>, \"input\"<\/span>: \"location\"<\/span>}}\n    )\n\n\nclass<\/span> Output<\/span>(BaseModel<\/span>):\n    # \u5b9a\u4e49\u8f93\u51fa BaseModel \u5305\u542b\u5b57\u6bb5 output<\/span>\n    output: str<\/span>\n\n\n# \u5c06\u8be5\u914d\u7f6e\u7684agent_executor\u6dfb\u52a0\u5230\u5e94\u7528\u7a0b\u5e8fapp\u7684\u8def\u7531\u4e2d\uff0c\u8def\u5f84\u4e3a \"\/agent\"<\/span>\nadd_routes(\n    app,\n    # agent_executor\u914d\u7f6e\u4e3a\u4f7f\u7528\u7279\u5b9a\u7684\u8f93\u5165\u548c\u8f93\u51fa\u7c7b\u578b<\/span>\n    agent_executor.with_types(input_type=Input, output_type=Output),\n    path=\"\/agent\"<\/span>,\n)\n\nif<\/span> __name__ == \"__main__\"<\/span>:\n    # \u5bfc\u5165 uvicorn \u6a21\u5757<\/span>\n    # uvicorn\u662f\u7528\u4e8e ASGI \u5e94\u7528\u7a0b\u5e8f\u7684\u8f7b\u91cf\u7ea7 Web \u670d\u52a1\u5668<\/span>\n    import<\/span> uvicorn\n\n    # \u8fd0\u884c\u4e3b\u5e94\u7528\u7a0b\u5e8f app\uff0c\u6307\u5b9a\u4e3b\u673a\u4e3a localhost\uff0c\u7aef\u53e3\u4e3a 8000<\/span>\n    uvicorn.run(app, host=\"localhost\"<\/span>, port=8000<\/span>)\n\n<\/code><\/pre>\n\n
\u542f\u52a8\u670d\u52a1<\/h2>\n
\u6267\u884c\u8fd9\u4e2a\u6587\u4ef6\u542f\u52a8\u670d\u52a1\uff0c\u5e76\u5728localhost:8000\u4e0a\u63d0\u4f9b\u670d\u52a1<\/p>\n\n
python serve.py\n<\/code><\/pre>\n\n\n
INFO:     Started server process [18352<\/span>]\nINFO:     Waiting for<\/span> application startup.\nINFO:     Application startup complete.\nINFO:     Uvicorn running on http:\/\/localhost:8000<\/span> (Press CTRL+C to quit)\n\n __          ___      .__   __.   _______      _______. _______ .______     ____    ____  _______\n|  |        \/   \\     |  \\ |  |  \/  _____|    \/       ||   ____||   _  \\    \\   \\  \/   \/ |   ____|\n|  |       \/  ^  \\    |   \\|  | |  |  __     |   (----`|  |__   |  |_)  |    \\   \\\/   \/  |  |__\n|  |      \/  \/_\\  \\   |  . `  | |  | |_ |     \\   \\    |   __|  |      \/      \\      \/   |   __|\n|  `----.\/  _____  \\  |  |\\   | |  |__| | .----)   |   |  |____ |  |\\  \\----.  \\    \/    |  |____\n|_______\/__\/     \\__\\ |__| \\__|  \\______| |_______\/    |_______|| _| `._____|   \\__\/     |_______|\n\nLANGSERVE: Playground for<\/span> chain \"\/agent\/\"<\/span> is<\/span> live at:\nLANGSERVE:  \u2502\nLANGSERVE:  \u2514\u2500\u2500> \/agent\/playground\/\nLANGSERVE:\nLANGSERVE: See all<\/span> available routes at \/docs\/\n<\/code><\/pre>\n\n
\u670d\u52a1\u4ea4\u4e92<\/h2>\n
\u6bcf\u4e2a LangServe \u670d\u52a1\u90fd\u5e26\u6709\u4e00\u4e2a\u7b80\u5355\u7684\u5185\u7f6e UI\uff0c\u7528\u4e8e\u914d\u7f6e\u548c\u8c03\u7528\u5177\u6709\u6d41\u8f93\u51fa\u548c\u4e2d\u95f4\u6b65\u9aa4\u53ef\u89c1\u6027\u7684\u5e94\u7528\u7a0b\u5e8f\u3002<\/p>\n
\u8bbf\u95ee\uff1ahttp:\/\/localhost:8000\/agent\/playground\/<\/code>\n\n
\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/figure>\n\n\u8bbe\u7f6e\u4e00\u4e2a\u5ba2\u6237\u7aef\uff0c\u4ee5\u4fbf\u4ee5\u7f16\u7a0b\u65b9\u5f0f\u4e0e\u6211\u4eec\u7684\u670d\u52a1\u8fdb\u884c\u4ea4\u4e92\u3002<\/p>\n\n
from<\/span> langserve import<\/span> RemoteRunnable\n\nremote_chain = RemoteRunnable(\"http:\/\/localhost:8000\/agent\/\"<\/span>)\nres = remote_chain.invoke({\n    \"input\"<\/span>: \"\u6210\u90fd\u4eca\u5929\u5929\u6c14\u60c5\u51b5\u600e\u6837?\"<\/span>,\n    \"chat_history\"<\/span>: []\n})\nprint<\/span>(res)\n<\/code><\/pre>\n","protected":false},"excerpt":{"rendered":"
LangChain\u662f\u4e00\u4e2a\u57fa\u4e8e\u5927\u8bed\u8a00\u6a21\u578b\u7528\u4e8e\u6784\u5efa\u7aef\u5230\u7aef\u8bed\u8a00\u6a21\u578b\u5e94\u7528\u7684\u6846\u67b6\uff0c\u5b83\u63d0\u4f9b\u4e86\u4e00\u7cfb\u5217\u5de5\u5177\u3001\u5957\u4ef6\u548c\u63a5\u53e3\uff0c\u8ba9\u5f00\u53d1\u8005\u4f7f\u7528\u8bed\u8a00\u6a21\u578b\u6765\u5b9e\u73b0\u5404\u79cd\u590d\u6742\u7684\u4efb\u52a1\uff0c\u5982\u6587\u672c\u5230\u56fe\u50cf\u7684\u751f\u6210\u3001\u6587\u6863\u95ee\u7b54\u3001\u804a\u5929\u673a\u5668\u4eba\u7b49\u3002<\/p>\n","protected":false},"author":1,"featured_media":0,"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":"3","footnotes":""},"categories":[3],"tags":[126,127,128,129,136],"collection":[],"class_list":["post-1967","post","type-post","status-publish","format-standard","hentry","category-fenlei2","tag-gpt","tag-ai","tag-128","tag-129","tag-136"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1967","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=1967"}],"version-history":[{"count":0,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1967\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media?parent=1967"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/categories?post=1967"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/tags?post=1967"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/collection?post=1967"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}