{"id":1613,"date":"2026-03-23T16:24:16","date_gmt":"2026-03-23T08:24:16","guid":{"rendered":"http:\/\/www.cmd137blog.top\/?p=1613"},"modified":"2026-03-23T17:38:13","modified_gmt":"2026-03-23T09:38:13","slug":"langchain-milvus-rag-%ef%bc%9apapertutor%e8%ae%ba%e6%96%87%e5%8a%a9%e6%95%99","status":"publish","type":"post","link":"http:\/\/www.cmd137blog.top\/?p=1613","title":{"rendered":"LangChain + Milvus RAG \uff1aPaperTutor\u8bba\u6587\u52a9\u6559"},"content":{"rendered":"\n<h1 class=\"wp-block-heading\">0 milvus\u524d\u63d0\u77e5\u8bc6\uff1a<\/h1>\n\n\n\n<p>\u672c\u6587\u53ea\u8986\u76d6\u672c\u9879\u76ee\u9700\u8981\u7684 Milvus \u57fa\u7840\u77e5\u8bc6\uff0c\u91cd\u70b9\u662f\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Milvus \u5728 RAG \u4e2d\u7684\u89d2\u8272<\/li>\n\n\n\n<li><code>langchain_milvus<\/code>&nbsp;\u63d0\u4f9b\u7684\u4e3b\u8981\u62bd\u8c61<\/li>\n\n\n\n<li>\u9700\u8981\u7406\u89e3\u7684\u6838\u5fc3\u6982\u5ff5\u3001API \u548c\u53c2\u6570<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"1-milvus-\u7684\u89d2\u8272\">1. Milvus \u7684\u89d2\u8272<\/h2>\n\n\n\n<p>Milvus \u662f\u5411\u91cf\u6570\u636e\u5e93\u3002<\/p>\n\n\n\n<p>\u5728 RAG \u4e2d\uff0c\u5b83\u8d1f\u8d23\u4e24\u4ef6\u4e8b\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u5b58\u50a8\u6587\u6863 chunk \u7684\u5411\u91cf<\/li>\n\n\n\n<li>\u6839\u636e\u95ee\u9898\u5411\u91cf\u68c0\u7d22\u6700\u76f8\u8fd1\u7684 chunk<\/li>\n<\/ol>\n\n\n\n<p>\u6700\u5c0f\u94fe\u8def\u53ef\u4ee5\u5199\u6210\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\"><code>\u6587\u6863 -> \u5207\u5757 -> embedding -> \u5b58\u5165 Milvus<br>\u95ee\u9898 -> embedding -> \u5728 Milvus \u68c0\u7d22 -> \u8fd4\u56de\u76f8\u5173 chunk<br><\/code><\/pre>\n\n\n\n<p>Milvus \u4e0d\u8d1f\u8d23\u751f\u6210\u56de\u7b54\uff0c\u5b83\u53ea\u8d1f\u8d23\uff1a<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u5b58\u5411\u91cf\u3001\u627e\u76f8\u4f3c\u5411\u91cf<\/p>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"2-\u6838\u5fc3\u6982\u5ff5\">2. \u6838\u5fc3\u6982\u5ff5<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"21-collection\">2.1 Collection<\/h3>\n\n\n\n<p><code>Collection<\/code>&nbsp;\u53ef\u4ee5\u7406\u89e3\u6210 Milvus \u4e2d\u7684\u4e00\u5f20\u5411\u91cf\u8868\u3002<\/p>\n\n\n\n<p>\u4e00\u6761\u8bb0\u5f55\u901a\u5e38\u5bf9\u5e94\u4e00\u4e2a chunk\uff0c\u8bb0\u5f55\u4e2d\u9664\u4e86\u5411\u91cf\uff0c\u8fd8\u4f1a\u5305\u542b\u6587\u672c\u548c metadata\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"22-vector\">2.2 Vector<\/h3>\n\n\n\n<p>\u5411\u91cf\u662f embedding \u6a21\u578b\u628a\u6587\u672c\u8f6c\u6362\u540e\u7684\u7ed3\u679c\u3002<\/p>\n\n\n\n<p>Milvus \u4e0d\u7406\u89e3\u539f\u59cb\u81ea\u7136\u8bed\u8a00\uff0c\u4f46\u53ef\u4ee5\u6bd4\u8f83\u5411\u91cf\u4e4b\u95f4\u7684\u76f8\u4f3c\u5ea6\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"23-dimension\">2.3 Dimension<\/h3>\n\n\n\n<p>\u7ef4\u5ea6\u5c31\u662f\u5411\u91cf\u957f\u5ea6\uff0c\u4f8b\u5982\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>768<\/code><\/li>\n\n\n\n<li><code>1024<\/code><\/li>\n<\/ul>\n\n\n\n<p>\u540c\u4e00\u4e2a collection \u7684\u5411\u91cf\u7ef4\u5ea6\u5fc5\u987b\u4e00\u81f4\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"24-search\">2.4 Search<\/h3>\n\n\n\n<p><code>Search<\/code>&nbsp;\u8868\u793a\u7ed9\u5b9a\u4e00\u4e2a query \u5411\u91cf\uff0c\u8fd4\u56de\u6700\u76f8\u8fd1\u7684\u82e5\u5e72\u6761\u8bb0\u5f55\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"25-top-k\">2.5 Top-K<\/h3>\n\n\n\n<p><code>top-k<\/code>&nbsp;\u8868\u793a\u8fd4\u56de\u524d\u591a\u5c11\u6761\u6700\u76f8\u5173\u7ed3\u679c\u3002<\/p>\n\n\n\n<p>\u5728 LangChain \u4e2d\u901a\u5e38\u5bf9\u5e94\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>search_kwargs={\"k\": ...}<\/code><\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"3-langchain_milvus-\u7684\u4f4d\u7f6e\">3.&nbsp;<code>langchain_milvus<\/code>&nbsp;\u7684\u4f4d\u7f6e<\/h2>\n\n\n\n<p><code>langchain_milvus<\/code>&nbsp;\u662f LangChain \u5bf9 Milvus \u7684\u5c01\u88c5\u5c42\u3002<\/p>\n\n\n\n<p>\u5173\u7cfb\u53ef\u4ee5\u7406\u89e3\u6210\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>Milvus \u6570\u636e\u5e93\n    \u2191\nlangchain_milvus\n    \u2191\nVectorStore \/ Retriever\n<\/code><\/code><\/pre>\n\n\n\n<p>\u5b83\u7684\u4f5c\u7528\u4e0d\u662f\u66ff\u4ee3 Milvus\uff0c\u800c\u662f\u628a Milvus \u63a5\u6210 LangChain \u53ef\u76f4\u63a5\u4f7f\u7528\u7684\u63a5\u53e3\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"4-langchain_milvus-\u4e3b\u8981-api\">4.&nbsp;<code>langchain_milvus<\/code>&nbsp;\u4e3b\u8981 API<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"41-milvus\">4.1&nbsp;<code>Milvus(...)<\/code><\/h3>\n\n\n\n<p>\u8fd9\u662f\u5411\u91cf\u5e93\u5bf9\u8c61\u672c\u8eab\u3002<\/p>\n\n\n\n<p>\u5178\u578b\u5199\u6cd5\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>vector_store = Milvus(\n    embedding_function=embedding,\n    collection_name=...,\n    connection_args=...,\n)\n<\/code><\/code><\/pre>\n\n\n\n<p>\u4f5c\u7528\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8fde\u63a5 Milvus<\/li>\n\n\n\n<li>\u7ed1\u5b9a embedding<\/li>\n\n\n\n<li>\u5173\u8054\u6307\u5b9a collection<\/li>\n\n\n\n<li>\u63d0\u4f9b LangChain \u98ce\u683c\u7684\u5199\u5165\u4e0e\u68c0\u7d22\u63a5\u53e3<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"42-add_documents\">4.2&nbsp;<code>add_documents(...)<\/code><\/h3>\n\n\n\n<p>\u7528\u4e8e\u5199\u5165&nbsp;<code>Document<\/code>&nbsp;\u5217\u8868\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>vector_store.add_documents(chunks)\n<\/code><\/code><\/pre>\n\n\n\n<p>\u80cc\u540e\u4f1a\u5b8c\u6210\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u63d0\u53d6\u6587\u672c<\/li>\n\n\n\n<li>\u63d0\u53d6 metadata<\/li>\n\n\n\n<li>\u751f\u6210\u5411\u91cf<\/li>\n\n\n\n<li>\u5199\u5165 Milvus<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"43-similarity_search\">4.3&nbsp;<code>similarity_search(...)<\/code><\/h3>\n\n\n\n<p>\u6309\u6587\u672c\u95ee\u9898\u76f4\u63a5\u505a\u76f8\u4f3c\u5ea6\u68c0\u7d22\u3002<\/p>\n\n\n\n<p>\u5b83\u5185\u90e8\u4f1a\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5148\u628a\u95ee\u9898\u505a embedding<\/li>\n\n\n\n<li>\u518d\u641c\u7d22 Milvus<\/li>\n\n\n\n<li>\u6700\u540e\u628a\u7ed3\u679c\u8f6c\u56de&nbsp;<code>Document<\/code><\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"44-as_retriever\">4.4&nbsp;<code>as_retriever(...)<\/code><\/h3>\n\n\n\n<p>\u628a vector store \u5305\u88c5\u6210 LangChain&nbsp;<code>Retriever<\/code>\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>retriever = vector_store.as_retriever(search_kwargs={\"k\": 4})\n<\/code><\/code><\/pre>\n\n\n\n<p>\u8fd9\u662f LangChain RAG \u4ee3\u7801\u91cc\u6700\u5e38\u89c1\u7684\u4e00\u6b65\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"45-retrieverinvoke\">4.5&nbsp;<code>Retriever.invoke(...)<\/code><\/h3>\n\n\n\n<p>\u6267\u884c\u68c0\u7d22\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>documents = retriever.invoke(question)\n<\/code><\/code><\/pre>\n\n\n\n<p>\u8f93\u5165\u662f\u95ee\u9898\uff0c\u8f93\u51fa\u662f\u76f8\u5173&nbsp;<code>Document<\/code>&nbsp;\u5217\u8868\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"5-\u5173\u952e\u53c2\u6570\">5. \u5173\u952e\u53c2\u6570<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"51-collection_name\">5.1&nbsp;<code>collection_name<\/code><\/h3>\n\n\n\n<p>\u51b3\u5b9a\u64cd\u4f5c\u7684\u662f\u54ea\u5f20 collection\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"52-embedding_function\">5.2&nbsp;<code>embedding_function<\/code><\/h3>\n\n\n\n<p>\u51b3\u5b9a\u6587\u672c\u5982\u4f55\u8f6c\u6362\u6210\u5411\u91cf\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"53-embedding_dim\">5.3&nbsp;<code>EMBEDDING_DIM<\/code><\/h3>\n\n\n\n<p>\u51b3\u5b9a\u671f\u671b\u7684\u5411\u91cf\u7ef4\u5ea6\u3002<\/p>\n\n\n\n<p>\u5fc5\u987b\u548c embedding \u5b9e\u9645\u8f93\u51fa\u4e00\u81f4\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"54-search_kwargsk-\">5.4&nbsp;<code>search_kwargs={\"k\": ...}<\/code><\/h3>\n\n\n\n<p>\u51b3\u5b9a\u6bcf\u6b21\u68c0\u7d22\u8fd4\u56de\u591a\u5c11\u6761\u7ed3\u679c\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"55-metric_type\">5.5&nbsp;<code>metric_type<\/code><\/h3>\n\n\n\n<p>\u76f8\u4f3c\u5ea6\u5ea6\u91cf\u65b9\u5f0f\u3002<\/p>\n\n\n\n<p>\u6587\u672c embedding \u573a\u666f\u5e38\u7528\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>COSINE<\/code><\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"6-\u5e38\u89c1\u95ee\u9898\">6. \u5e38\u89c1\u95ee\u9898<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"61-\u5411\u91cf\u7ef4\u5ea6\u4e0d\u4e00\u81f4\">6.1 \u5411\u91cf\u7ef4\u5ea6\u4e0d\u4e00\u81f4<\/h3>\n\n\n\n<p>\u5982\u679c\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u914d\u7f6e\u7ef4\u5ea6<\/li>\n\n\n\n<li>embedding \u5b9e\u9645\u8f93\u51fa\u7ef4\u5ea6<\/li>\n\n\n\n<li>collection schema \u7ef4\u5ea6<\/li>\n<\/ul>\n\n\n\n<p>\u4e09\u8005\u4e0d\u4e00\u81f4\uff0c\u5c31\u4f1a\u51fa\u9519\uff0c\u6216\u8005\u5efa\u51fa\u9519\u8bef\u7684 collection\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"62-collection-\u590d\u7528\">6.2 collection \u590d\u7528<\/h3>\n\n\n\n<p>\u5982\u679c\u590d\u7528\u65e7 collection\uff0c\u4f46\u914d\u7f6e\u5df2\u7ecf\u53d8\u5316\uff0c\u5bb9\u6613\u51fa\u73b0\u51b2\u7a81\u3002<\/p>\n\n\n\n<p>\u6d4b\u8bd5\u4e0d\u540c\u7ef4\u5ea6\u6216\u4e0d\u540c\u5b9e\u9a8c\u65f6\uff0c\u66f4\u5b89\u5168\u7684\u65b9\u5f0f\u662f\u4f7f\u7528\u65b0\u7684 collection \u540d\u79f0\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"63-\u9ad8\u5c42\u5c01\u88c5\u7684\u9ed1\u76d2\u611f\">6.3 \u9ad8\u5c42\u5c01\u88c5\u7684\u9ed1\u76d2\u611f<\/h3>\n\n\n\n<p><code>langchain_milvus<\/code>&nbsp;\u8ba9\u4ee3\u7801\u66f4\u77ed\uff0c\u4f46\u4e5f\u610f\u5473\u7740\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5e95\u5c42\u7ec6\u8282\u88ab\u5305\u8d77\u6765\u4e86<\/li>\n\n\n\n<li>\u51fa\u9519\u65f6\u4e0d\u5982\u539f\u751f\u65b9\u5f0f\u76f4\u89c2<\/li>\n<\/ul>\n\n\n\n<p>\u5b83\u7b80\u5316\u7684\u662f\u63a5\u5165\u65b9\u5f0f\uff0c\u4e0d\u662f\u5e95\u5c42\u539f\u7406\u3002<\/p>\n\n\n\n<h1 class=\"wp-block-heading\">LangChain + Milvus RAG \u9879\u76ee\u5b66\u4e60\u6307\u5357<\/h1>\n\n\n\n<p>\u672c\u9879\u76ee\u7528\u4e8e\u5b66\u4e60\u4e00\u4e2a\u6700\u5c0f\u53ef\u8fd0\u884c\u7684 LangChain RAG \u7cfb\u7edf\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4f7f\u7528&nbsp;<code>PyPDFLoader<\/code>&nbsp;\u8bfb\u53d6 PDF<\/li>\n\n\n\n<li>\u4f7f\u7528&nbsp;<code>RecursiveCharacterTextSplitter<\/code>&nbsp;\u5207\u5206\u6587\u6863<\/li>\n\n\n\n<li>\u4f7f\u7528 DashScope embedding \u751f\u6210\u5411\u91cf<\/li>\n\n\n\n<li>\u4f7f\u7528 Milvus \u5b58\u50a8\u548c\u68c0\u7d22\u5411\u91cf<\/li>\n\n\n\n<li>\u4f7f\u7528 LangChain&nbsp;<code>Retriever<\/code>&nbsp;+ Prompt + LLM \u7ec4\u7ec7\u95ee\u7b54<\/li>\n<\/ul>\n\n\n\n<p>\u5f53\u524d\u63a8\u8350\u5b66\u4e60\u5206\u652f\uff1a<\/p>\n\n\n\n<p><code>experiment\/langchain-milvus-retriever<\/code><\/p>\n\n\n\n<p>\u8fd9\u6761\u5206\u652f\u7684\u91cd\u70b9\u662f\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7528&nbsp;<code>langchain-milvus<\/code>&nbsp;\u7406\u89e3 LangChain \u98ce\u683c\u7684\u5411\u91cf\u5e93\u63a5\u5165<\/li>\n\n\n\n<li>\u7528&nbsp;<code>VectorStore<\/code>&nbsp;\/&nbsp;<code>Retriever<\/code>&nbsp;\u7406\u89e3\u6807\u51c6 RAG \u94fe\u8def<\/li>\n\n\n\n<li>\u5728\u4fdd\u7559 Milvus \u7684\u540c\u65f6\uff0c\u89c2\u5bdf LangChain \u5c01\u88c5\u5c42\u5230\u5e95\u5305\u88c5\u4e86\u4ec0\u4e48<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"1-\u5b66\u4e60\u76ee\u6807\">1. \u5b66\u4e60\u76ee\u6807<\/h2>\n\n\n\n<p>\u9605\u8bfb\u548c\u8fd0\u884c\u672c\u9879\u76ee\u65f6\uff0c\u5e94\u91cd\u70b9\u7406\u89e3\u8fd9\u4e9b\u95ee\u9898\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>RAG \u4e0e\u76f4\u63a5\u8c03\u7528\u5927\u6a21\u578b\u7684\u533a\u522b\u662f\u4ec0\u4e48<\/li>\n\n\n\n<li>\u4e3a\u4ec0\u4e48\u957f PDF \u9700\u8981\u5207\u5206\u6210 chunk<\/li>\n\n\n\n<li><code>chunk_size<\/code>&nbsp;\u4e0e&nbsp;<code>chunk_overlap<\/code>&nbsp;\u4f1a\u5982\u4f55\u5f71\u54cd\u68c0\u7d22\u6548\u679c<\/li>\n\n\n\n<li>\u4ec0\u4e48\u662f embedding\uff0c\u4ec0\u4e48\u662f\u5411\u91cf\u7ef4\u5ea6<\/li>\n\n\n\n<li>Milvus \u4e2d&nbsp;<code>collection<\/code>\u3001<code>schema<\/code>\u3001<code>index<\/code>\u3001<code>search<\/code>&nbsp;\u5404\u81ea\u662f\u4ec0\u4e48<\/li>\n\n\n\n<li>LangChain \u4e2d&nbsp;<code>VectorStore<\/code>&nbsp;\u4e0e&nbsp;<code>Retriever<\/code>&nbsp;\u5404\u81ea\u8d1f\u8d23\u4ec0\u4e48<\/li>\n\n\n\n<li><code>langchain-milvus<\/code>&nbsp;\u4e0e\u539f\u751f&nbsp;<code>pymilvus<\/code>&nbsp;\u7684\u4ee3\u7801\u7ec4\u7ec7\u5dee\u5f02\u662f\u4ec0\u4e48<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"2-\u7cfb\u7edf\u5168\u666f\">2. \u7cfb\u7edf\u5168\u666f<\/h2>\n\n\n\n<p>\u9879\u76ee\u7684\u4e3b\u94fe\u8def\u5982\u4e0b\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>PDF\n  -&gt; Loader\n  -&gt; Documents\n  -&gt; Text Splitter\n  -&gt; Chunks\n  -&gt; Embedding\n  -&gt; Milvus\n  -&gt; Retriever\n  -&gt; Prompt\n  -&gt; LLM\n  -&gt; Answer\n<\/code><\/code><\/pre>\n\n\n\n<p>\u53ef\u4ee5\u518d\u62c6\u6210\u4e24\u9636\u6bb5\uff1a<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u9636\u6bb5-a\u5efa\u5e93\">\u9636\u6bb5 A\uff1a\u5efa\u5e93<\/h3>\n\n\n\n<p>\u628a\u8bba\u6587\u53d8\u6210\u53ef\u68c0\u7d22\u77e5\u8bc6\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u9636\u6bb5-b\u95ee\u7b54\">\u9636\u6bb5 B\uff1a\u95ee\u7b54<\/h3>\n\n\n\n<p>\u628a\u7528\u6237\u95ee\u9898\u8f6c\u6210 query\uff0c\u68c0\u7d22\u76f8\u5173 chunk\uff0c\u518d\u4ea4\u7ed9 LLM \u7ec4\u7ec7\u56de\u7b54\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"3-\u9879\u76ee\u7ed3\u6784\">3. \u9879\u76ee\u7ed3\u6784<\/h2>\n\n\n\n<pre class=\"wp-block-code\"><code><code>rag_app\/\n  app.py\n  config.py\n  ingestion.py\n  milvus_store.py\n  chatbot.py\n  prompts.py\n  dashscope_embeddings.py\n<\/code><\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"4-\u6a21\u5757\u8bf4\u660e\">4. \u6a21\u5757\u8bf4\u660e<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"rag_appapppy\"><code>rag_app\/app.py<\/code><\/h3>\n\n\n\n<p>\u9879\u76ee\u5165\u53e3\uff0c\u66b4\u9732\u4e24\u4e2a\u547d\u4ee4\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>ingest-papers<\/code><\/li>\n\n\n\n<li><code>chat<\/code><\/li>\n<\/ul>\n\n\n\n<p>\u4f5c\u7528\u662f\u63a5\u6536\u547d\u4ee4\u5e76\u8c03\u7528\u4e1a\u52a1\u6a21\u5757\uff0c\u4e0d\u627f\u62c5\u6838\u5fc3\u903b\u8f91\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"rag_appconfigpy\"><code>rag_app\/config.py<\/code><\/h3>\n\n\n\n<p>\u7edf\u4e00\u8bfb\u53d6&nbsp;<code>.env<\/code>&nbsp;\u914d\u7f6e\uff0c\u6574\u7406\u4e3a&nbsp;<code>RagSettings<\/code>\u3002<\/p>\n\n\n\n<p>\u91cd\u70b9\u914d\u7f6e\u5305\u62ec\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>EMBEDDING_MODEL<\/code><\/li>\n\n\n\n<li><code>EMBEDDING_DIM<\/code><\/li>\n\n\n\n<li><code>MILVUS_URI<\/code><\/li>\n\n\n\n<li><code>MILVUS_TOKEN<\/code><\/li>\n\n\n\n<li><code>MILVUS_DB<\/code><\/li>\n\n\n\n<li><code>MILVUS_COLLECTION<\/code><\/li>\n\n\n\n<li><code>TOP_K<\/code><\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"rag_appingestionpy\"><code>rag_app\/ingestion.py<\/code><\/h3>\n\n\n\n<p>\u5efa\u5e93\u4e3b\u6d41\u7a0b\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>load_documents\n  -&gt; split_documents\n  -&gt; add_documents\n<\/code><\/code><\/pre>\n\n\n\n<p>\u5b66\u4e60\u91cd\u70b9\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>LangChain&nbsp;<code>Document<\/code>&nbsp;\u7684\u7ed3\u6784<\/li>\n\n\n\n<li>chunk \u5207\u5206\u7684\u5fc5\u8981\u6027<\/li>\n\n\n\n<li>chunk overlap \u7684\u4f5c\u7528<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"rag_appmilvus_storepy\"><code>rag_app\/milvus_store.py<\/code><\/h3>\n\n\n\n<p>\u5c06 Milvus \u5305\u88c5\u4e3a LangChain \u53ef\u4f7f\u7528\u7684&nbsp;<code>VectorStore<\/code>\u3002<\/p>\n\n\n\n<p>\u5b66\u4e60\u91cd\u70b9\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Milvus \u672c\u4f53\u8d1f\u8d23 collection\u3001schema\u3001index\u3001search<\/li>\n\n\n\n<li>LangChain \u5c06\u5176\u62bd\u8c61\u4e3a&nbsp;<code>add_documents()<\/code>\u3001<code>similarity_search()<\/code>\u3001<code>as_retriever()<\/code><\/li>\n\n\n\n<li><code>langchain-milvus<\/code>&nbsp;\u5e76\u4e0d\u6539\u53d8 Milvus \u539f\u7406\uff0c\u53ea\u662f\u7edf\u4e00\u4e86\u8c03\u7528\u63a5\u53e3<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"rag_appchatbotpy\"><code>rag_app\/chatbot.py<\/code><\/h3>\n\n\n\n<p>\u95ee\u7b54\u4e3b\u94fe\uff0c\u5173\u952e\u7ed3\u6784\u662f\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>self.retriever = self.vector_store.as_retriever(\n    search_kwargs={\"k\": self.settings.k}\n)\ndocuments = self.retriever.invoke(question)\n<\/code><\/code><\/pre>\n\n\n\n<p>\u8fd9\u4f53\u73b0\u4e86\u5178\u578b RAG \u5206\u5de5\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>VectorStore<\/code>\uff1a\u5b58\u50a8\u4e0e\u641c\u7d22\u80fd\u529b<\/li>\n\n\n\n<li><code>Retriever<\/code>\uff1a\u6839\u636e\u95ee\u9898\u627e\u8bc1\u636e<\/li>\n\n\n\n<li><code>LLM<\/code>\uff1a\u9605\u8bfb\u8bc1\u636e\u5e76\u7ec4\u7ec7\u56de\u7b54<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"rag_apppromptspy\"><code>rag_app\/prompts.py<\/code><\/h3>\n\n\n\n<p>\u5b9a\u4e49\u63d0\u793a\u8bcd\u4e0e\u56de\u7b54\u89c4\u5219\u3002<\/p>\n\n\n\n<p>\u91cd\u70b9\u4e0d\u662f\u6587\u98ce\uff0c\u800c\u662f\u884c\u4e3a\u7ea6\u675f\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u57fa\u4e8e\u68c0\u7d22\u8bc1\u636e\u56de\u7b54<\/li>\n\n\n\n<li>\u4e0d\u786e\u5b9a\u65f6\u660e\u786e\u8bf4\u660e<\/li>\n\n\n\n<li>\u591a\u5b50\u95ee\u9898\u5206\u9879\u56de\u7b54<\/li>\n\n\n\n<li>\u5c3d\u91cf\u9644\u5e26\u6765\u6e90<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"rag_appdashscope_embeddingspy\"><code>rag_app\/dashscope_embeddings.py<\/code><\/h3>\n\n\n\n<p>\u5bf9 DashScope embedding \u505a\u9879\u76ee\u7ea7\u5c01\u88c5\u3002<\/p>\n\n\n\n<p>\u4e3b\u8981\u5904\u7406\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5411\u91cf\u7ef4\u5ea6\u63a7\u5236<\/li>\n\n\n\n<li>batch \u9650\u5236<\/li>\n\n\n\n<li>\u5b9e\u9645\u8fd4\u56de\u7ef4\u5ea6\u6821\u9a8c<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"5-\u6838\u5fc3\u6982\u5ff5\">5. \u6838\u5fc3\u6982\u5ff5<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"51-document\">5.1 Document<\/h3>\n\n\n\n<p>LangChain&nbsp;<code>Document<\/code>&nbsp;\u901a\u5e38\u5305\u542b\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>page_content<\/code><\/li>\n\n\n\n<li><code>metadata<\/code><\/li>\n<\/ul>\n\n\n\n<p>\u8fd9\u8868\u793a RAG \u5904\u7406\u7684\u4e0d\u662f\u7eaf\u6587\u672c\uff0c\u800c\u662f\u201c\u6587\u672c&nbsp;+&nbsp;\u6765\u6e90\u4fe1\u606f\u201d\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"52-chunk\">5.2 Chunk<\/h3>\n\n\n\n<p>\u957f\u6587\u6863\u4e0d\u80fd\u76f4\u63a5\u6574\u4f53\u505a\u4e00\u4e2a\u5411\u91cf\uff0c\u56e0\u6b64\u9700\u8981\u5207\u6210\u66f4\u5c0f\u7684 chunk\u3002<\/p>\n\n\n\n<p>\u5207\u5206\u7684\u539f\u56e0\u5305\u62ec\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8ba9\u68c0\u7d22\u66f4\u7cbe\u786e<\/li>\n\n\n\n<li>\u964d\u4f4e\u4e0a\u4e0b\u6587\u5197\u4f59<\/li>\n\n\n\n<li>\u8ba9\u95ee\u9898\u4e0e\u5c40\u90e8\u5185\u5bb9\u5339\u914d<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"53-embedding\">5.3 Embedding<\/h3>\n\n\n\n<p>embedding \u662f\u628a\u6587\u672c\u6620\u5c04\u6210\u5411\u91cf\u3002<\/p>\n\n\n\n<p>Milvus \u4e0d\u7406\u89e3\u81ea\u7136\u8bed\u8a00\uff0c\u4f46\u53ef\u4ee5\u6bd4\u8f83\u5411\u91cf\u4e4b\u95f4\u7684\u76f8\u4f3c\u5ea6\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"54-vector-dimension\">5.4 Vector Dimension<\/h3>\n\n\n\n<p>\u5411\u91cf\u7ef4\u5ea6\u662f schema \u7684\u4e00\u90e8\u5206\u3002<\/p>\n\n\n\n<p>\u5982\u679c collection \u7684\u5411\u91cf\u5b57\u6bb5\u6309\u67d0\u4e2a\u7ef4\u5ea6\u5efa\u597d\uff0c\u540e\u7eed\u5199\u5165\u7684\u5411\u91cf\u957f\u5ea6\u5fc5\u987b\u4e00\u81f4\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"55-vectorstore\">5.5 VectorStore<\/h3>\n\n\n\n<p><code>VectorStore<\/code>&nbsp;\u662f LangChain \u5bf9\u201c\u5411\u91cf\u5b58\u50a8&nbsp;+&nbsp;\u76f8\u4f3c\u5ea6\u641c\u7d22\u201d\u7684\u7edf\u4e00\u62bd\u8c61\u3002<\/p>\n\n\n\n<p>\u5728\u672c\u9879\u76ee\u4e2d\uff0c\u5e95\u5c42\u5b9e\u73b0\u662f Milvus\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"56-retriever\">5.6 Retriever<\/h3>\n\n\n\n<p><code>Retriever<\/code>&nbsp;\u662f\u4e13\u95e8\u8d1f\u8d23\u201c\u6839\u636e\u95ee\u9898\u627e\u76f8\u5173\u6587\u6863\u201d\u7684\u7ec4\u4ef6\u3002<\/p>\n\n\n\n<p>\u5178\u578b\u94fe\u8def\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>question -&gt; retriever -&gt; documents -&gt; llm -&gt; answer\n<\/code><\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"6-\u73af\u5883\u51c6\u5907\">6. \u73af\u5883\u51c6\u5907<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u5b89\u88c5\u4f9d\u8d56\">\u5b89\u88c5\u4f9d\u8d56<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code><code>python -m venv .venv\n.\\.venv\\Scripts\\activate\npip install -r requirements.txt\n<\/code><\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u5173\u952e\u73af\u5883\u53d8\u91cf\">\u5173\u952e\u73af\u5883\u53d8\u91cf<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code><code>DASHSCOPE_API_KEY=...\nBASE_URL=...\nMILVUS_URI=...\nMILVUS_TOKEN=...\nMILVUS_DB=paper_rag\nMILVUS_COLLECTION=paper_chunks\nEMBEDDING_MODEL=text-embedding-v4\nEMBEDDING_DIM=1024\n<\/code><\/code><\/pre>\n\n\n\n<p>\u8bf4\u660e\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>MILVUS_COLLECTION<\/code>&nbsp;\u51b3\u5b9a\u5199\u5165\u54ea\u5f20 collection<\/li>\n\n\n\n<li><code>EMBEDDING_DIM<\/code>&nbsp;\u5fc5\u987b\u4e0e\u5b9e\u9645\u8f93\u51fa\u5411\u91cf\u957f\u5ea6\u4e00\u81f4<\/li>\n\n\n\n<li>\u6d4b\u8bd5\u4e0d\u540c\u7ef4\u5ea6\u65f6\uff0c\u5efa\u8bae\u4f7f\u7528\u65b0\u7684 collection \u540d\u79f0<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"7-\u5e38\u7528\u547d\u4ee4\">7. \u5e38\u7528\u547d\u4ee4<\/h2>\n\n\n\n<pre class=\"wp-block-code\"><code><code>python -m rag_app.app --help\npython -m rag_app.app ingest-papers 2310.11511v1.pdf\npython -m rag_app.app chat\n<\/code><\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"8-\u5b66\u4e60\u987a\u5e8f\">8. \u5b66\u4e60\u987a\u5e8f<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u7b2c\u4e00\u904d\u8dd1\u901a\">\u7b2c\u4e00\u904d\uff1a\u8dd1\u901a<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code><code>python -m rag_app.app ingest-papers 2310.11511v1.pdf\npython -m rag_app.app chat\n<\/code><\/code><\/pre>\n\n\n\n<p>\u76ee\u6807\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u77e5\u9053\u4e00\u4e2a\u6700\u5c0f RAG \u9879\u76ee\u600e\u6837\u8fd0\u884c<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u7b2c\u4e8c\u904d\u6293\u4e3b\u7ebf\">\u7b2c\u4e8c\u904d\uff1a\u6293\u4e3b\u7ebf<\/h3>\n\n\n\n<p>\u91cd\u70b9\u770b\u8fd9\u51e0\u884c\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>vector_store.add_documents(chunks)\n<\/code><\/code><\/pre>\n\n\n\n<pre class=\"wp-block-code\"><code><code>self.retriever = self.vector_store.as_retriever(...)\n<\/code><\/code><\/pre>\n\n\n\n<pre class=\"wp-block-code\"><code><code>documents = self.retriever.invoke(question)\n<\/code><\/code><\/pre>\n\n\n\n<p>\u76ee\u6807\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7406\u89e3 LangChain \u98ce\u683c RAG \u7684\u6838\u5fc3\u63a5\u53e3<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u7b2c\u4e09\u904d\u505a\u5b9e\u9a8c\">\u7b2c\u4e09\u904d\uff1a\u505a\u5b9e\u9a8c<\/h3>\n\n\n\n<p>\u4f18\u5148\u8c03\u6574\u8fd9\u4e9b\u53c2\u6570\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>CHUNK_SIZE<\/code><\/li>\n\n\n\n<li><code>CHUNK_OVERLAP<\/code><\/li>\n\n\n\n<li><code>TOP_K<\/code><\/li>\n\n\n\n<li><code>EMBEDDING_DIM<\/code><\/li>\n\n\n\n<li><code>MILVUS_COLLECTION<\/code><\/li>\n<\/ul>\n\n\n\n<p>\u89c2\u5bdf\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u68c0\u7d22\u7ed3\u679c\u662f\u5426\u66f4\u96c6\u4e2d<\/li>\n\n\n\n<li>\u56de\u7b54\u662f\u5426\u66f4\u5177\u4f53<\/li>\n\n\n\n<li>\u662f\u5426\u66f4\u5bb9\u6613\u4e22\u5931\u4e0a\u4e0b\u6587<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"9-langchain-milvus-\u7684\u5178\u578b\u95ee\u9898\">9.&nbsp;<code>langchain-milvus<\/code>&nbsp;\u7684\u5178\u578b\u95ee\u9898<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"91-\u8fde\u63a5-alias-\u95ee\u9898\">9.1 \u8fde\u63a5 alias \u95ee\u9898<\/h3>\n\n\n\n<p><code>langchain-milvus<\/code>&nbsp;\u5185\u90e8\u6df7\u7528\u4e86&nbsp;<code>MilvusClient<\/code>&nbsp;\u4e0e ORM alias \u673a\u5236\u3002<\/p>\n\n\n\n<p>\u5728\u67d0\u4e9b\u7248\u672c\u7ec4\u5408\u4e0b\uff0c\u53ef\u80fd\u51fa\u73b0\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>ConnectionNotExistException: should create connection first.\n<\/code><\/code><\/pre>\n\n\n\n<p>\u8fd9\u7c7b\u95ee\u9898\u901a\u5e38\u4e0d\u662f\u51ed\u8bc1\u9519\u8bef\uff0c\u800c\u662f\u8fde\u63a5\u7ba1\u7406\u8854\u63a5\u95ee\u9898\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"92-\u9ad8\u5c42\u5c01\u88c5\u66f4\u9ed1\u76d2\">9.2 \u9ad8\u5c42\u5c01\u88c5\u66f4\u9ed1\u76d2<\/h3>\n\n\n\n<p>\u5f53\u4ee3\u7801\u5199\u6210\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>vector_store.add_documents(chunks)\n<\/code><\/code><\/pre>\n\n\n\n<p>\u5e95\u5c42\u4ecd\u7136\u4f1a\u53d1\u751f\u5f88\u591a\u4e8b\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6587\u672c\u4e0e metadata \u63d0\u53d6<\/li>\n\n\n\n<li>embedding \u8c03\u7528<\/li>\n\n\n\n<li>collection \u68c0\u67e5\u6216\u521d\u59cb\u5316<\/li>\n\n\n\n<li>\u5199\u5165\u4e0e\u7d22\u5f15\u64cd\u4f5c<\/li>\n<\/ul>\n\n\n\n<p>\u4f18\u70b9\u662f\u4ee3\u7801\u77ed\uff0c\u7f3a\u70b9\u662f\u51fa\u9519\u65f6\u4e0d\u900f\u660e\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"93-\u591a\u5c42\u4f9d\u8d56\u7684\u517c\u5bb9\u6027\">9.3 \u591a\u5c42\u4f9d\u8d56\u7684\u517c\u5bb9\u6027<\/h3>\n\n\n\n<p>\u94fe\u8def\u6d89\u53ca\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>langchain<\/code><\/li>\n\n\n\n<li><code>langchain-core<\/code><\/li>\n\n\n\n<li><code>langchain-community<\/code><\/li>\n\n\n\n<li><code>langchain-milvus<\/code><\/li>\n\n\n\n<li><code>pymilvus<\/code><\/li>\n<\/ul>\n\n\n\n<p>\u5c42\u6570\u8d8a\u591a\uff0c\u517c\u5bb9\u6027\u98ce\u9669\u8d8a\u9ad8\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"10-\u4ee3\u7801\u5bf9\u7167langchain-milvus-\u5305\u88c5\u4e86\u4ec0\u4e48\">10. \u4ee3\u7801\u5bf9\u7167\uff1a<code>langchain-milvus<\/code>&nbsp;\u5305\u88c5\u4e86\u4ec0\u4e48<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"101-\u5efa\u5e93\u5199\u6cd5\">10.1 \u5efa\u5e93\u5199\u6cd5<\/h3>\n\n\n\n<h4 class=\"wp-block-heading\" id=\"langchain-\u98ce\u683c\">LangChain \u98ce\u683c<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code><code>vector_store = get_vector_store(embeddings, settings)\nvector_store.add_documents(chunks)\n<\/code><\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\" id=\"\u539f\u751f-pymilvus\">\u539f\u751f&nbsp;<code>pymilvus<\/code><\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code><code>texts = &#91;chunk.page_content for chunk in chunks]\nvectors = embeddings.embed_documents(texts)\n\nrows = &#91;]\nfor chunk, vector in zip(chunks, vectors, strict=True):\n    rows.append(\n        {\n            \"text\": chunk.page_content,\n            \"source\": chunk.metadata.get(\"source\", \"unknown\"),\n            \"page\": chunk.metadata.get(\"page\", -1),\n            \"vector\": vector,\n        }\n    )\n\ncollection.insert(rows)\ncollection.flush()\n<\/code><\/code><\/pre>\n\n\n\n<p><code>langchain-milvus<\/code>&nbsp;\u5728\u8fd9\u91cc\u5305\u88c5\u6389\u4e86\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4ece&nbsp;<code>Document<\/code>&nbsp;\u63d0\u53d6\u6587\u672c\u548c metadata<\/li>\n\n\n\n<li>embedding \u8c03\u7528\u8854\u63a5<\/li>\n\n\n\n<li>\u5199\u5165\u7ed3\u6784\u8f6c\u6362<\/li>\n\n\n\n<li>\u63d2\u5165\u6d41\u7a0b\u5c01\u88c5<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"102-\u68c0\u7d22\u5199\u6cd5\">10.2 \u68c0\u7d22\u5199\u6cd5<\/h3>\n\n\n\n<h4 class=\"wp-block-heading\" id=\"langchain-\u98ce\u683c-1\">LangChain \u98ce\u683c<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code><code>retriever = vector_store.as_retriever(search_kwargs={\"k\": 4})\ndocuments = retriever.invoke(question)\n<\/code><\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\" id=\"\u539f\u751f-pymilvus-1\">\u539f\u751f&nbsp;<code>pymilvus<\/code><\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code><code>query_vector = embedding.embed_query(question)\nresults = collection.search(\n    data=&#91;query_vector],\n    anns_field=\"vector\",\n    param={\"metric_type\": \"COSINE\", \"params\": {}},\n    limit=4,\n    output_fields=&#91;\"text\", \"source\", \"page\"],\n)\n<\/code><\/code><\/pre>\n\n\n\n<p>\u5e76\u624b\u52a8\u8f6c\u6362\u4e3a\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><code>documents = &#91;]\nfor result in results&#91;0]:\n    entity = result.entity\n    documents.append(\n        Document(\n            page_content=entity.get(\"text\"),\n            metadata={\n                \"source\": entity.get(\"source\", \"unknown\"),\n                \"page\": entity.get(\"page\", \"?\"),\n            },\n        )\n    )\n<\/code><\/code><\/pre>\n\n\n\n<p><code>langchain-milvus<\/code>&nbsp;\u5728\u8fd9\u91cc\u5305\u88c5\u6389\u4e86\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>query embedding \u4e0e\u641c\u7d22\u8854\u63a5<\/li>\n\n\n\n<li>\u641c\u7d22\u7ed3\u679c\u89e3\u6790<\/li>\n\n\n\n<li>\u7ed3\u679c\u8f6c&nbsp;<code>Document<\/code><\/li>\n\n\n\n<li><code>Retriever<\/code>&nbsp;\u63a5\u53e3\u62bd\u8c61<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"103-\u7ed3\u8bba\">10.3 \u7ed3\u8bba<\/h3>\n\n\n\n<p><code>langchain-milvus<\/code>&nbsp;\u5305\u88c5\u7684\u662f\u63a5\u5165\u7ec6\u8282\uff0c\u4e0d\u662f RAG \u539f\u7406\u3002<\/p>\n\n\n\n<p>\u5b83\u5e2e\u52a9\u7b80\u5316\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6587\u6863\u8f6c\u6362<\/li>\n\n\n\n<li>\u5411\u91cf\u5199\u5165\u8854\u63a5<\/li>\n\n\n\n<li>\u68c0\u7d22\u7ed3\u679c\u8f6c\u6362<\/li>\n\n\n\n<li>retriever \u63a5\u53e3<\/li>\n<\/ul>\n\n\n\n<p>\u5b83\u4e0d\u4f1a\u66ff\u4ee3\u8fd9\u4e9b\u57fa\u7840\u95ee\u9898\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5411\u91cf\u7ef4\u5ea6\u4e00\u81f4\u6027<\/li>\n\n\n\n<li>chunk \u8bbe\u8ba1<\/li>\n\n\n\n<li>top-k \u9009\u62e9<\/li>\n\n\n\n<li>embedding \u8d28\u91cf<\/li>\n\n\n\n<li>\u4f9d\u8d56\u517c\u5bb9\u6027<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"11-dashscope-\u7ef4\u5ea6\u53c2\u6570\u8bf4\u660e\">11. DashScope \u7ef4\u5ea6\u53c2\u6570\u8bf4\u660e<\/h2>\n\n\n\n<p>DashScope embedding \u6709\u4e24\u6761\u5e38\u89c1\u8c03\u7528\u8def\u5f84\uff1a<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"openai-\u517c\u5bb9\u63a5\u53e3\">OpenAI \u517c\u5bb9\u63a5\u53e3<\/h3>\n\n\n\n<p>\u53c2\u6570\u540d\uff1a<\/p>\n\n\n\n<p><code>dimensions<\/code><\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"dashscope-sdk-\u539f\u751f\u63a5\u53e3\">DashScope SDK \u539f\u751f\u63a5\u53e3<\/h3>\n\n\n\n<p>\u53c2\u6570\u540d\uff1a<\/p>\n\n\n\n<p><code>dimension<\/code><\/p>\n\n\n\n<p>\u672c\u9879\u76ee\u5f53\u524d\u8d70\u7684\u662f&nbsp;<code>dashscope.TextEmbedding.call(...)<\/code>&nbsp;\u8fd9\u6761\u8def\u5f84\uff0c\u56e0\u6b64\u5fc5\u987b\u4f7f\u7528\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><span style=\"background-color: initial; font-family: inherit; font-size: inherit; text-align: initial; color: initial;\">dimension=768<\/span><\/code><\/pre>\n\n\n\n<p>\u5982\u679c\u8bef\u5199\u6210\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code><span style=\"background-color: rgba(0, 0, 0, 0.2); font-family: inherit; font-size: inherit; text-align: initial; color: initial;\">dimensions=768<\/span><\/code><\/pre>\n\n\n\n<p>\u53ef\u80fd\u4f1a\u88ab\u9759\u9ed8\u5ffd\u7565\uff0c\u5e76\u56de\u9000\u5230\u6a21\u578b\u9ed8\u8ba4\u7ef4\u5ea6\u3002<\/p>\n\n\n\n<h1 class=\"wp-block-heading\">\u6e90\u7801\uff1a<\/h1>\n\n\n\n<p><a href=\"https:\/\/github.com\/CMD137\/PaperTutor\">CMD137\/PaperTutor<\/a><\/p>\n\n\n\n<h1 class=\"wp-block-heading\">\u7ed3\u679c\u793a\u4f8b\uff1a<\/h1>\n\n\n\n<pre class=\"wp-block-code\"><code>(.venv) C:\\Users\\29924\\Desktop\\developtest\\milvusTest&gt;python -m rag_app.app ingest-papers 2310.11511v1.pdf\nStarting ingestion...\nResolved input paths: &#91;WindowsPath('2310.11511v1.pdf')]\nLoading PDF: 2310.11511v1.pdf\nLoaded 30 pages from 2310.11511v1.pdf\nLoaded 30 source pages\nSplit into 150 chunks\nEmbedding config: {'model': 'text-embedding-v4', 'configured_dim': 768, 'embedding_object_dim': 768}\nProbe document embedding dimension: 768\nAdding chunk documents through langchain-milvus...\nStored 150 chunks into paper_chunks\nStored 150 chunks inside collection paper_chunks.\n\n(.venv) C:\\Users\\29924\\Desktop\\developtest\\milvusTest&gt;python -m rag_app.app chat\nPaper tutor ready. Type 'exit' to stop.\n\nYou: \u8bb2\u8bb2self-rag\n\nTimings:\n- retrieval: 0.39s\n- context_format: 0.00s\n- llm_generation: 43.91s\n- total: 44.31s\n\nTutor:\n\u7528\u6237\u95ee\u9898\u53ea\u5305\u542b\u4e00\u4e2a\u5b50\u95ee\u9898\uff0c\u5373\u201c\u8bb2\u8bb2self-rag\u201d\u3002\n\n### \u95ee\u98981: \u8bb2\u8bb2SELF-RAG\n\n**\u7ed3\u8bba\uff1a**\nSELF-RAG\u662f\u4e00\u79cd\u6539\u8fdb\u7684\u8bed\u8a00\u6a21\u578b\uff08LM\uff09\uff0c\u5b83\u901a\u8fc7\u5728\u751f\u6210\u8fc7\u7a0b\u4e2d\u5f15\u5165\u81ea\u6211\u53cd\u601d\u673a\u5236\u6765\u63d0\u5347\u6587\u672c\u751f\u6210\u7684\u8d28\u91cf\u3001\u4e8b\u5b9e\u6027\u548c\u53ef\u9a8c\u8bc1\u6027\u3002\n\n**\u89e3\u91ca\uff1a**\n1. **\u81ea\u6211\u53cd\u601d\u673a\u5236\uff1a** SELF-RAG\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u5728\u751f\u6210\u4efb\u52a1\u8f93\u51fa\u7684\u540c\u65f6\uff0c\u751f\u6210\u4e00\u4e9b\u7279\u6b8a\u7684\u6807\u8bb0\uff08\u79f0\u4e3a\u53cd\u5c04\u6807\u8bb0\u6216reflection tokens\uff09\uff0c\u8fd9\u4e9b\u6807\u8bb0\u53ef\u4ee5\u5206\u4e3a\u68c0\u7d22\u6807\u8bb0\u548c\u6279\u8bc4\u6807\u8bb0\u3002\u68c0\u7d22\u6807\u8bb0\u7528\u4e8e\u6307\u793a\u9700\u8981\u68c0\u7d22\u5916\u90e8\u4fe1\u606f\uff0c\u800c\u6279\u8bc4\u6807\u8bb0\u5219\u7528\u4e8e\u8bc4\u4f30\u751f\u6210\u5185\u5bb9\u7684\u8d28\u91cf &#91;source: 2310.11511v1.pdf p1]\u3002\n\n2. **\u6309\u9700\u68c0\u7d22\u4e0e\u5904\u7406\uff1a** \u5f53\u6a21\u578b\u5224\u65ad\u51fa\u7ee7\u7eed\u751f\u6210\u65f6\u9700\u8981\u8865\u5145\u5916\u90e8\u4fe1\u606f\u65f6\uff0c\u4f1a\u8f93\u51fa\u4e00\u4e2a\u68c0\u7d22\u6807\u8bb0\uff0c\u89e6\u53d1\u68c0\u7d22\u5668\u6a21\u578b\u53bb\u83b7\u53d6\u76f8\u5173\u4fe1\u606f\u3002\u4e4b\u540e\uff0cSELF-RAG\u80fd\u591f\u540c\u65f6\u5904\u7406\u591a\u4e2a\u68c0\u7d22\u5230\u7684\u6bb5\u843d\uff0c\u8bc4\u4f30\u5b83\u4eec\u7684\u76f8\u5173\u6027\uff0c\u5e76\u57fa\u4e8e\u6b64\u751f\u6210\u6700\u7ec8\u7684\u4efb\u52a1\u8f93\u51fa &#91;source: 2310.11511v1.pdf p1]\u3002\u8fd9\u79cd\u65b9\u6cd5\u76f8\u6bd4\u4f20\u7edf\u7684\u68c0\u7d22\u589e\u5f3a\u751f\u6210\u65b9\u6cd5\u66f4\u52a0\u7075\u6d3b\uff0c\u56e0\u4e3a\u540e\u8005\u901a\u5e38\u65e0\u8bba\u662f\u5426\u5fc5\u8981\u90fd\u4f1a\u56fa\u5b9a\u5730\u68c0\u7d22\u4e00\u5b9a\u6570\u91cf\u7684\u6587\u6863 &#91;source: 2310.11511v1.pdf p2]\u3002\n\n3. **\u63d0\u9ad8\u8d28\u91cf\u548c\u53ef\u8ffd\u6eaf\u6027\uff1a** SELF-RAG\u4e0d\u4ec5\u589e\u5f3a\u4e86\u751f\u6210\u6587\u672c\u7684\u4e8b\u5b9e\u51c6\u786e\u6027\uff0c\u8fd8\u63d0\u4f9b\u4e86\u6bcf\u4e2a\u751f\u6210\u90e8\u5206\u7684\u5f15\u7528\u6765\u6e90\u53ca\u5176\u81ea\u6211\u8bc4\u4f30\u7ed3\u679c\uff0c\u4f7f\u5f97\u66f4\u5bb9\u6613\u8fdb\u884c\u4e8b\u5b9e\u9a8c\u8bc1 &#91;source: 2310.11511v1.pdf p2]\u3002\u6b64\u5916\uff0c\u901a\u8fc7\u7ec6\u7c92\u5ea6\u7684\u81ea\u6211\u53cd\u601d\u8fc7\u7a0b\uff0c\u8be5\u6a21\u578b\u80fd\u591f\u9009\u62e9\u6700\u4f73\u53ef\u80fd\u7684\u8f93\u51fa\uff0c\u4ece\u800c\u63d0\u9ad8\u4e86\u751f\u6210\u5185\u5bb9\u7684\u6574\u4f53\u8d28\u91cf &#91;source: 2310.11511v1.pdf p4]\u3002\n\n**\u5c40\u9650\uff1a**\n\u6839\u636e\u5f53\u524d\u63d0\u4f9b\u7684\u4e0a\u4e0b\u6587\uff0c\u6ca1\u6709\u63d0\u5230SELF-RAG\u7684\u5177\u4f53\u5b9e\u73b0\u7ec6\u8282\u5982\u4f7f\u7528\u7684\u6570\u636e\u96c6\u540d\u79f0\u3001\u786e\u5207\u7684\u8bad\u7ec3\u6b65\u9aa4\u7b49\u6280\u672f\u7ec6\u8282\u3002\n\n**\u8ffd\u95ee\u5efa\u8bae\uff1a**\n- SELF-RAG\u662f\u5982\u4f55\u5177\u4f53\u5b9e\u73b0\u5bf9\u68c0\u7d22\u5230\u7684\u4fe1\u606f\u8fdb\u884c\u76f8\u5173\u6027\u8bc4\u4ef7\u7684\uff1f\n- \u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0cSELF-RAG\u76f8\u6bd4\u5176\u4ed6RAG\u65b9\u6cd5\u6709\u54ea\u4e9b\u5177\u4f53\u7684\u6027\u80fd\u4f18\u52bf\uff1f\n<\/code><\/pre>\n","protected":false},"excerpt":{"rendered":"<p>0 milvus\u524d\u63d0\u77e5\u8bc6\uff1a \u672c\u6587\u53ea\u8986\u76d6\u672c\u9879\u76ee\u9700\u8981\u7684 Milvus \u57fa\u7840\u77e5\u8bc6\uff0c\u91cd\u70b9\u662f\uff1a 1. Milvus \u7684\u89d2 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[82,66],"tags":[86,87,85],"class_list":["post-1613","post","type-post","status-publish","format-standard","hentry","category-82","category-66","tag-langchain","tag-milvus","tag-rag"],"_links":{"self":[{"href":"http:\/\/www.cmd137blog.top\/index.php?rest_route=\/wp\/v2\/posts\/1613","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/www.cmd137blog.top\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.cmd137blog.top\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.cmd137blog.top\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/www.cmd137blog.top\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1613"}],"version-history":[{"count":19,"href":"http:\/\/www.cmd137blog.top\/index.php?rest_route=\/wp\/v2\/posts\/1613\/revisions"}],"predecessor-version":[{"id":1635,"href":"http:\/\/www.cmd137blog.top\/index.php?rest_route=\/wp\/v2\/posts\/1613\/revisions\/1635"}],"wp:attachment":[{"href":"http:\/\/www.cmd137blog.top\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1613"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.cmd137blog.top\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1613"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.cmd137blog.top\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1613"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}