{"id":10021508,"date":"2026-10-27T09:41:01","date_gmt":"2026-10-27T01:41:01","guid":{"rendered":"https:\/\/itmax.cn\/?p=10021508"},"modified":"2026-10-27T09:41:16","modified_gmt":"2026-10-27T01:41:16","slug":"pageindex-vectorless-rag-20261027","status":"publish","type":"post","link":"https:\/\/itmax.cn\/?p=10021508","title":{"rendered":"\u5411\u91cf\u5e93\u6551\u4e0d\u4e86\u957f\u6587\u6863\uff5cPageIndex \u7684\u6811\u7d22\u5f15\uff0c\u6211\u88c5\u5b8c\u8bd5\u5230\u6458\u8981\u90a3\u6b65\u5c31\u505c\u4e86"},"content":{"rendered":"<p>\u4e1a\u52a1\u540c\u4e8b\u95ee\u6211\u7684\u95ee\u9898\uff0c\u6765\u6765\u56de\u56de\u5c31\u90a3\u4e48\u51e0\u53e5\uff1a\u300c\u5bf9\u65b9\u903e\u671f\u4ed8\u6b3e\uff0c\u6211\u4eec\u7684\u8fdd\u7ea6\u91d1\u600e\u4e48\u7b97\u300d\u300c\u8fd9\u6761\u6392\u4ed6\u6761\u6b3e\u662f\u4e0d\u662f\u628a\u6574\u4e2a\u534e\u4e1c\u90fd\u5708\u8fdb\u53bb\u4e86\u300d\u300c\u9a8c\u6536\u4e0d\u5408\u683c\u7684\u8865\u6551\u671f\u9650\u5199\u7684\u51e0\u5929\u300d\u3002<\/p>\n<p>\u8fd9\u4e9b\u95ee\u9898\u7684\u7b54\u6848\uff0c\u5168\u5728\u5408\u540c PDF \u91cc\uff0c\u800c\u4e14\u5168\u5728\u67d0\u4e00\u8282\u7684\u67d0\u4e00\u6bb5\u3002\u6211\u505a\u8fc7\u7684\u4e8b\u5f88\u81ea\u7136\uff1a\u628a\u5408\u540c\u704c\u8fdb\u77e5\u8bc6\u5e93\uff0c\u5207\u5757\u3001\u8f6c\u5411\u91cf\u3001\u68c0\u7d22\u3001\u585e\u7ed9\u6a21\u578b\u3002\u7136\u540e\u5c31\u5361\u5728\u8fd9\u91cc\u4e86\u2014\u2014\u95ee\u8fdd\u7ea6\u91d1\uff0c\u53ec\u56de\u6765\u7684\u516b\u6bb5\u91cc\u6709\u4e03\u6bb5\u8bb2\u7684\u662f\u4ed8\u6b3e\u8282\u70b9\u548c\u8d26\u671f\uff0c\u771f\u6b63\u90a3\u53e5\u300c\u903e\u671f\u4e09\u5341\u65e5\u4ee5\u4e0a\uff0c\u6309\u672a\u4ed8\u6b3e\u9879\u65e5\u4e07\u5206\u4e4b\u4e94\u8ba1\u300d\u538b\u6839\u6ca1\u8fdb\u5019\u9009\u96c6\u3002\u539f\u56e0\u4e0d\u590d\u6742\uff1a\u90a3\u53e5\u8bdd\u7684\u4e0a\u4e00\u53e5\u662f\u300c\u8fdd\u7ea6\u8d23\u4efb\u300d\uff0c\u518d\u4e0a\u4e00\u53e5\u662f\u5c0f\u6807\u9898\u300c\u7b2c\u4e5d\u6761 \u8fdd\u7ea6\u4e0e\u6551\u6d4e\u300d\u3002\u5207\u7247\u6309\u5b57\u6570\u786c\u5207\uff0c\u300c\u7b2c\u4e5d\u6761\u300d\u88ab\u5207\u5230\u4e86\u4e0a\u4e00\u5757\uff0c\u300c\u903e\u671f\u4e09\u5341\u65e5\u4ee5\u4e0a\u300d\u88ab\u5207\u5230\u4e86\u4e0b\u4e00\u5757\uff0c\u4e2d\u95f4\u90a3\u5757\u5355\u72ec\u770b\uff0c\u8bed\u4e49\u4e0a\u8ddf\u300c\u4ed8\u6b3e\u65b9\u5f0f\u300d\u6ca1\u4ec0\u4e48\u533a\u522b\u3002<\/p>\n<p>\u8fd9\u4e0d\u662f\u67d0\u4e00\u6b21\u7684\u610f\u5916\uff0c\u662f\u6bcf\u4e2a\u6708\u90fd\u4f1a\u78b0\u4e0a\u7684\u4e8b\u3002\u6211\u8bd5\u8fc7\u7684\u8865\u4e01\u4e5f\u90fd\u8bd5\u8fc7\uff1a\u628a chunk \u4ece\u516b\u767e\u5b57\u52a0\u5230\u4e00\u5343\u4e94\u3001\u91cd\u53e0\u4e24\u767e\u5b57\u3001\u52a0 reranker\u3001\u52a0\u5173\u952e\u8bcd\u53ec\u56de\u4e00\u8def\u6df7\u68c0\u3002\u6bcf\u52a0\u4e00\u5c42\uff0c\u53ec\u56de\u7387\u662f\u4f1a\u4e0a\u53bb\u4e00\u70b9\uff0c\u4ee3\u4ef7\u662f\u566a\u58f0\u4e5f\u4e00\u8d77\u4e0a\u53bb\uff0c\u6a21\u578b\u5f00\u59cb\u5728\u4e03\u516b\u6bb5\u4f3c\u662f\u800c\u975e\u7684\u6761\u6b3e\u4e4b\u95f4\u7f16\u7b54\u6848\u3002\u6df7\u68c0\u8c03\u5230\u80fd\u7528\u7684\u7a0b\u5ea6\uff0c\u6211\u5df2\u7ecf\u5728\u7ef4\u62a4\u4e00\u5957\u6bd4\u6211\u4e1a\u52a1\u4ee3\u7801\u8fd8\u590d\u6742\u7684\u68c0\u7d22\u94fe\u8def\u4e86\u3002<\/p>\n<p>\u6240\u4ee5\u6211\u770b\u5230 PageIndex \u90a3\u53e5 slogan \u7684\u65f6\u5019\uff0c\u662f\u88ab\u6233\u4e86\u4e00\u4e0b\u7684\uff1a<strong>similarity \u2260 relevance<\/strong>\u3002\u76f8\u4f3c\u4e0d\u7b49\u4e8e\u76f8\u5173\uff0c\u8fd9\u8bdd\u5728\u5408\u540c\u3001\u8d22\u62a5\u3001\u62db\u80a1\u4e66\u8fd9\u7c7b\u6587\u6863\u4e0a\u662f\u771f\u7684\u3002<\/p>\n<h2 id=\"_1\">\u5b83\u628a\u6587\u6863\u76ee\u5f55\u5f53\u6210\u7d22\u5f15\u7528\uff0c\u8fd9\u6761\u4e0d\u662f\u5671\u5934<\/h2>\n<p>\u601d\u8def\u672c\u8eab\u5f88\u6734\u7d20\uff0c\u6734\u7d20\u5230\u6709\u70b9\u53cd\u5e38\uff1a\u65e2\u7136\u4eba\u6709\u76ee\u5f55\uff0c\u4e3a\u4ec0\u4e48\u4e0d\u62ff\u76ee\u5f55\u5f53\u7d22\u5f15\u3002<\/p>\n<p>\u5efa\u6811\u662f\u4e00\u6b65\uff0c\u68c0\u7d22\u662f\u53e6\u4e00\u6b65\u3002\u5efa\u6811\u7684\u65f6\u5019\uff0c\u5b83\u4ece PDF \u7684\u7248\u9762\u91cc\u628a\u6807\u9898\u5c42\u7ea7\u62bd\u51fa\u6765\uff0c\u505a\u6210\u4e00\u4e2a\u5e26 <code>node_id<\/code>\u3001\u9875\u7801\u533a\u95f4\u3001\u7236\u5b50\u548c\u5144\u5f1f\u5173\u7cfb\u7684 JSON\uff1b\u68c0\u7d22\u7684\u65f6\u5019\uff0c\u6a21\u578b\u62ff\u7740\u95ee\u9898\u5728\u8fd9\u68f5\u6811\u4e0a\u8d70\uff0c\u5224\u65ad\u8be5\u8fdb\u54ea\u4e2a\u5206\u652f\uff0c\u8d70\u5230\u53f6\u5b50\u8282\u70b9\u518d\u53bb\u8bfb\u90a3\u51e0\u9875\u7684\u6b63\u6587\u3002\u6574\u4e2a\u8fc7\u7a0b\u8ddf\u4eba\u7ffb\u4e00\u672c\u539a\u62a5\u544a\u6ca1\u533a\u522b\uff1a\u5148\u770b\u76ee\u5f55\uff0c\u518d\u7ffb\u5230\u7b2c\u4e5d\u6761\uff0c\u518d\u8bfb\u90a3\u4e09\u9875\u3002<\/p>\n<p>\u771f\u6b63\u80fd\u8dd1\u7684\u5165\u53e3\u6709\u4e09\u4e2a\uff0c\u6211\u6328\u4e2a\u770b\u8fc7\uff1a<\/p>\n<pre><code class=\"language-bash\"># \u4ed3\u5e93\u81ea\u5e26\u811a\u672c\uff0cPDF \u8d70 flash \u6a21\u5f0f\uff08\u9ed8\u8ba4\uff09\npython run_pageindex.py --pdf_path .\/contract.pdf\n\n# Markdown \u76f4\u63a5\u8fdb\uff0c\u4e0d\u9700\u8981\u6a21\u578b\u53c2\u4e0e\u5c31\u80fd\u51fa\u6811\npython run_pageindex.py --md_path .\/notes.md\n\n# SDK\uff1a\u672c\u5730\u5efa\u6811\uff0c\u81ea\u5df1\u7684\u6a21\u578b\u56de\u7b54\npython -c &quot;\nfrom pageindex import PageIndexClient\nc = PageIndexClient(index='gpt-6-luna', chat='gpt-6-sol')\ndoc_id = c.submit_document('contract.pdf')['doc_id']\nprint(c.chat('\u903e\u671f\u4ed8\u6b3e\u7684\u8fdd\u7ea6\u91d1\u600e\u4e48\u7b97', doc_id=doc_id))\n&quot;\n<\/code><\/pre>\n<p>\u5de5\u5177\u5c42\u662f\u56db\u4e2a\uff1a<code>browse_documents<\/code> \u5217\u6e05\u5355\u3001<code>get_document_structure<\/code> \u53d6\u6811\u3001<code>get_page_content<\/code> \u6309\u9875\u53d6\u6b63\u6587\u3001<code>remove_document<\/code> \u5220\u3002SDK \u91cc\u8fd8\u6709 <code>as_openai_tools()<\/code> \u548c <code>as_anthropic_tools()<\/code> \u628a\u8fd9\u4e00\u7ec4\u5305\u88c5\u6210\u4e3b\u6d41\u6846\u67b6\u8ba4\u7684\u5de5\u5177\u5b9a\u4e49\u3002\u8fd9\u51e0\u4e2a\u63a5\u53e3\u662f\u5b9e\u6253\u5b9e\u5199\u5728\u4ee3\u7801\u91cc\u7684\uff0c\u4e0d\u662f\u5ba3\u4f20\u9875 PPT\u3002<\/p>\n<p>\u5efa\u6811\u7684\u7c97\u7ec6\u7a0b\u5ea6\u662f\u53ef\u4ee5\u62e7\u7684\uff0c\u65cb\u94ae\u90fd\u5728 <code>config.yaml<\/code> \u91cc\uff0c\u9ed8\u8ba4\u503c\u633a\u4fdd\u5b88\uff1a<\/p>\n<pre><code class=\"language-yaml\">toc_check_page_num: 20        # \u53ea\u5728\u524d 20 \u9875\u91cc\u627e\u76ee\u5f55\nmax_page_num_each_node: 10    # \u4e00\u4e2a\u8282\u70b9\u6700\u591a 10 \u9875\nmax_token_num_each_node: 20000\nif_add_node_summary: &quot;yes&quot;    # \u6bcf\u4e2a\u8282\u70b9\u8ba9\u6a21\u578b\u5199\u4e00\u6bb5\u6458\u8981\n<\/code><\/pre>\n<p>\u524d\u4e24\u4e2a\u6570\u5b57\u51b3\u5b9a\u4e86\u6811\u7684\u5206\u8fa8\u7387\u3002\u4e00\u4efd\u4e24\u767e\u9875\u7684\u6587\u6863\u6309\u300c\u6bcf\u8282\u70b9\u6700\u591a\u5341\u9875\u300d\u5207\uff0c\u6700\u591a\u5206\u51fa\u4e8c\u5341\u4e2a\u53f6\u5b50\uff1b\u5408\u540c\u8fd9\u79cd\u4e5d\u6761\u5341\u4e8c\u6b3e\u7684\u6587\u6863\uff0c\u53f6\u5b50\u4f1a\u7c97\u5230\u4e00\u6761\u4e00\u8282\u4e00\u5c42\u90fd\u5e26\u7740\uff0c\u8ddf\u6211\u60f3\u8981\u7684\u300c\u7cbe\u786e\u5b9a\u4f4d\u5230\u90a3\u53e5\u8fdd\u7ea6\u6761\u6b3e\u300d\u8fd8\u5dee\u4e00\u622a\u3002\u8981\u66f4\u7ec6\u5c31\u5f97\u628a <code>max_page_num_each_node<\/code> \u5f80\u4e0b\u8c03\uff0c\u4ee3\u4ef7\u662f\u8282\u70b9\u53d8\u591a\u3001\u6458\u8981\u8c03\u7528\u53d8\u591a\u3001\u68c0\u7d22\u65f6\u6a21\u578b\u8981\u770b\u7684\u6811\u4e5f\u53d8\u957f\u3002\u8fd9\u6761\u6ca1\u6709\u6807\u51c6\u7b54\u6848\uff0c\u53ea\u80fd\u62ff\u81ea\u5df1\u7684\u6587\u6863\u8bd5\u3002<\/p>\n<p>\u6458\u8981\u90a3\u5757\u8fd8\u6709\u4e2a\u7ec6\u8282\uff1a\u6bcf\u4e2a\u8282\u70b9\u7684\u6458\u8981\u9ed8\u8ba4\u538b\u5230 150 \u8bcd\uff08<code>SUMMARY_MAX_WORDS = 150<\/code>\uff09\uff0c\u53d1\u7ed9\u6a21\u578b\u7684 reasoning effort \u662f <code>none<\/code>\u3002\u5b98\u65b9\u5728 README \u91cc\u7684\u8bf4\u6cd5\u662f\u300c\u5efa\u6811\u7528\u57fa\u7840\u6a21\u578b\u5c31\u591f\uff0c\u771f\u6b63\u5403\u80fd\u529b\u7684\u662f\u68c0\u7d22\u65f6\u7684 chat \u6a21\u578b\u300d\u3002\u6211\u8ba4\u540c\u8fd9\u4e2a\u5224\u65ad\u7684\u52a8\u673a\u2014\u2014\u6811\u7684\u7ed3\u6784\u6765\u81ea\u7248\u9762\uff0c\u6a21\u578b\u53ea\u8d1f\u8d23\u5f52\u7eb3\u2014\u2014\u4f46 150 \u8bcd\u7684\u6458\u8981\u53bb\u63cf\u8ff0\u4e00\u4efd\u8d22\u62a5\u91cc\u4e00\u8282\u4e94\u9875\u7684\u6b63\u6587\uff0c\u4fe1\u606f\u635f\u8017\u662f\u514d\u4e0d\u6389\u7684\uff0c\u8fd9\u4e5f\u662f\u6211\u6000\u7591\u68c0\u7d22\u51c6\u786e\u7387\u4f1a\u6389\u7684\u4e00\u4e2a\u70b9\uff0c\u53ea\u662f\u6ca1 key\uff0c\u9a8c\u4e0d\u4e86\u3002<\/p>\n<p>\u54ea\u4e9b\u662f\u771f\u80fd\u7528\uff0c\u54ea\u4e9b\u8fd8\u53ea\u662f\u8bdd\u672f\uff0c\u6211\u5206\u5f00\u8bf4\uff1a<\/p>\n<table>\n<thead>\n<tr>\n<th>\u80fd\u529b<\/th>\n<th>\u6211\u7684\u5224\u65ad<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u79bb\u7ebf\u5efa\u6811\uff08PDF\u3001Markdown\uff09<\/td>\n<td>\u771f\u80fd\u7528\uff0c\u6211\u81ea\u5df1\u8dd1\u901a\u4e86\uff0c\u4e0d\u9700\u8981\u4efb\u4f55 key<\/td>\n<\/tr>\n<tr>\n<td>SDK \u672c\u5730\u6a21\u5f0f\uff08\u5efa\u6811 + \u68c0\u7d22\uff09<\/td>\n<td>\u80fd\u7528\uff0c\u4f46\u6458\u8981\u548c\u68c0\u7d22\u4e24\u6b65\u5fc5\u987b\u6709\u6a21\u578b key<\/td>\n<\/tr>\n<tr>\n<td>\u9875\u7801\u7ea7\u5f15\u7528<\/td>\n<td>\u672c\u5730\u6a21\u5f0f\u6709\uff0c\u80fd\u8ffd\u5230\u7b2c\u51e0\u9875<\/td>\n<\/tr>\n<tr>\n<td>\u5757\u7ea7\u5f15\u7528\u3001OCR\u3001\u56fe\u7247\u7406\u89e3<\/td>\n<td>\u53ea\u5728\u4e91\u7aef\uff0c\u672c\u5730\u7248\u6ca1\u6709<\/td>\n<\/tr>\n<tr>\n<td>MCP server<\/td>\n<td>\u53ea\u6709\u4e91\u7aef\u7248\uff0c\u672c\u5730\u4e0d\u8d77<\/td>\n<\/tr>\n<tr>\n<td>98.7% FinanceBench<\/td>\n<td>\u90a3\u662f Mafin 2.5 \u7684\u6210\u7ee9\u5355\uff0c\u4e0d\u662f <code>pip install<\/code> \u5b8c\u5c31\u6709\u7684\u80fd\u529b<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u6700\u540e\u4e00\u884c\u6211\u60f3\u591a\u8bf4\u4e00\u53e5\u3002\u90a3\u4e2a 98.7% \u51fa\u81ea\u4ed6\u4eec\u81ea\u5bb6\u7684 Mafin 2.5 \u9879\u76ee\uff0c\u8dd1\u7684\u662f\u5b8c\u6574 FinanceBench \u9898\u5e93\uff0c\u5c5e\u4e8e\u81ea\u5bb6\u7cfb\u7edf\u5728\u81ea\u5bb6\u57fa\u51c6\u4e0a\u7684\u6210\u7ee9\u3002\u771f\u6b63\u5bf9\u5f00\u6e90\u7248\u6709\u53c2\u8003\u4ef7\u503c\u7684\u662f\u53e6\u4e00\u4e2a\u4ed3\u5e93\u2014\u2014<code>PageIndex-OSS-Benchmark<\/code>\uff0c\u7528 34 \u4efd PDF\u30011,945 \u9875\u300162 \u9053\u9898\u7684\u672c\u5730 flash \u7d22\u5f15\u6765\u8dd1\uff0c\u4ed6\u4eec\u8fd8\u5199\u660e\u4e86\u4e00\u4ef6\u4e8b\uff1a\u7d22\u5f15\u4e0d\u8fdb\u53bb\u7684\u6587\u6863\u88ab\u6392\u9664\u5728\u5916\uff0c\u800c\u4e0d\u662f\u7b97\u4f5c\u5931\u8d25\u3002\u8fd9\u53e5\u8bdd\u5f88\u5173\u952e\uff0c\u5b83\u610f\u5473\u7740\u90a3\u4e2a\u51c6\u786e\u7387\u53ea\u63cf\u8ff0\u300cPageIndex \u80fd\u5403\u8fdb\u53bb\u7684\u6587\u6863\u300d\uff0c\u4e0d\u63cf\u8ff0\u4f60\u7684\u6587\u6863\u3002<\/p>\n<h2 id=\"_2\">\u7701\u6389\u5411\u91cf\u5e93\uff0c\u8d26\u5355\u632a\u5230\u4e86\u68c0\u7d22\u90a3\u4e00\u523b<\/h2>\n<p>\u8fd9\u662f\u6211\u6700\u60f3\u63b0\u626f\u7684\u5730\u65b9\uff0c\u56e0\u4e3a\u5927\u90e8\u5206\u63a8\u6587\u53ea\u8bb2\u524d\u534a\u53e5\u3002<\/p>\n<table>\n<thead>\n<tr>\n<th><\/th>\n<th>\u5411\u91cf\u68c0\u7d22<\/th>\n<th>\u6811\u7d22\u5f15 + \u63a8\u7406<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u5efa\u7d22\u5f15<\/td>\n<td>\u4e00\u6b21\u6027 embedding\uff0c\u5feb\u3001\u4fbf\u5b9c<\/td>\n<td>\u904d\u5386\u7248\u9762 + \u6bcf\u4e2a\u8282\u70b9\u4e00\u6b21\u6a21\u578b\u8c03\u7528<\/td>\n<\/tr>\n<tr>\n<td>\u5355\u6b21\u68c0\u7d22<\/td>\n<td>\u4e00\u6b21\u5411\u91cf\u6bd4\u5bf9\uff0c\u6beb\u79d2\u7ea7<\/td>\n<td>\u6a21\u578b\u6cbf\u6811\u8d70\uff0c\u51e0\u6b21\u5230\u5341\u51e0\u6b21\u8c03\u7528<\/td>\n<\/tr>\n<tr>\n<td>\u51fa\u9519\u5f62\u6001<\/td>\n<td>\u53ec\u56de\u4e0d\u76f8\u5173\u4f46\u76f8\u4f3c\u7684\u7247\u6bb5<\/td>\n<td>\u8d70\u9519\u5206\u652f\u3001\u8bfb\u9519\u8282\u70b9\u3001\u6458\u8981\u6f0f\u4fe1\u606f<\/td>\n<\/tr>\n<tr>\n<td>\u53ef\u8ffd\u6eaf\u6027<\/td>\n<td>\u76f8\u4f3c\u5ea6\u5206\u6570\uff0c\u8bf4\u4e0d\u6e05\u4e3a\u4ec0\u4e48<\/td>\n<td>\u8282\u70b9 id + \u9875\u7801\uff0c\u80fd\u753b\u8def\u5f84<\/td>\n<\/tr>\n<tr>\n<td>\u6269\u5c55\u6027<\/td>\n<td>\u767e\u4e07\u7ea7\u5411\u91cf\u5f88\u6210\u719f<\/td>\n<td>\u5355\u6587\u6863\u5f88\u5f3a\uff0c\u8de8\u6587\u6863\u8981\u989d\u5916\u4e00\u5c42<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u7701\u6389\u5411\u91cf\u5e93\u7701\u6389\u7684\u662f\u8fd0\u7ef4\uff0c\u4e0d\u662f\u94b1\u3002\u539f\u6765\u4f60\u4ed8\u7684\u662f Pinecone \u6216\u8005 Milvus \u7684\u8d26\u5355\uff0c\u73b0\u5728\u4ed8\u7684\u662f\u6a21\u578b token \u7684\u8d26\u5355\u3002\u5b98\u65b9\u5f00\u6e90\u57fa\u51c6\u91cc 12 \u7ec4\u914d\u7f6e\u7684\u6570\u5b57\u6446\u5f97\u5f88\u6e05\u695a\uff0c\u6211\u6311\u56db\u884c\uff1a<\/p>\n<table>\n<thead>\n<tr>\n<th>chat \u6a21\u578b<\/th>\n<th>reasoning effort<\/th>\n<th>62 \u9898\u7b54\u5bf9<\/th>\n<th>\u5355\u6b21\u67e5\u8be2\u5747\u4ef7<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>gpt-5.6-luna<\/td>\n<td>none<\/td>\n<td>53<\/td>\n<td>$0.0031<\/td>\n<\/tr>\n<tr>\n<td>gpt-5.6-luna<\/td>\n<td>high<\/td>\n<td>60<\/td>\n<td>$0.0036<\/td>\n<\/tr>\n<tr>\n<td>gpt-5.6-terra<\/td>\n<td>medium<\/td>\n<td>61<\/td>\n<td>$0.0303<\/td>\n<\/tr>\n<tr>\n<td>gpt-5.6-sol<\/td>\n<td>medium<\/td>\n<td>62<\/td>\n<td>$0.0810<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u4fbf\u5b9c\u6a21\u578b\u516b\u4e94\u6298\u7684\u51c6\u786e\u7387\uff0c\u4e00\u6b21\u4e09\u5206\u94b1\uff1b\u8d35\u6a21\u578b\u5168\u5bf9\uff0c\u4e00\u6b21\u516b\u5206\u94b1\uff0c\u4e2d\u95f4\u5dee\u4e86\u4e8c\u5341\u516d\u500d\u3002\u800c\u540c\u4e00\u6b21\u67e5\u8be2\u7528\u5411\u91cf\u5e93\uff0c\u6210\u672c\u51e0\u4e4e\u53ef\u4ee5\u5ffd\u7565\u4e0d\u8ba1\u3002\u6240\u4ee5\u8fd9\u7b14\u4ea4\u6613\u7684\u672c\u8d28\u662f\uff1a<strong>\u7528\u5ef6\u8fdf\u548c token \u6362\u51c6\u786e\u7387\u548c\u53ef\u89e3\u91ca\u6027<\/strong>\u3002\u5b83\u4e0d\u662f\u514d\u8d39\u5348\u9910\uff0c\u5b83\u662f\u53e6\u4e00\u79cd\u4ed8\u8d39\u65b9\u5f0f\u3002<\/p>\n<p>\u4ee3\u4ef7\u5728\u56db\u79cd\u60c5\u51b5\u4e0b\u4f1a\u53d8\u5f97\u4e0d\u53ef\u63a5\u53d7\uff0c\u6211\u4e00\u6761\u6761\u5212\u51fa\u6765\uff1a<\/p>\n<p><strong>\u9ad8\u5e76\u53d1\u5728\u7ebf\u95ee\u7b54\u4e0d\u884c\u3002<\/strong> \u6bcf\u6b21\u68c0\u7d22\u90fd\u8981\u6a21\u578b\u6765\u56de\u8d70\u6811\uff0c\u793e\u533a\u91cc\u5b9e\u6d4b\u7684\u91cf\u7ea7\u662f\u5355\u6b21\u67e5\u8be2\u51e0\u5341\u79d2\u5230\u4e00\u4e24\u5206\u949f\uff0c\u6ca1\u6709\u7f13\u5b58\u7684\u60c5\u51b5\u3002\u8fd9\u5bf9\u4e00\u4e2a\u804a\u5929\u7a97\u53e3\u662f\u707e\u96be\uff0c\u5bf9\u4e00\u4e2a\u5206\u6790\u5e08\u81ea\u5df1\u7528\u7684\u5de5\u5177\u5c31\u5b8c\u5168\u53ef\u4ee5\u63a5\u53d7\u3002\u770b\u7684\u662f\u573a\u666f\uff0c\u4e0d\u662f\u6570\u5b57\u3002<\/p>\n<p><strong>\u788e\u7247\u5316\u8bed\u6599\u4e0d\u884c\u3002<\/strong> \u5b83\u7684\u6811\u6765\u81ea\u6587\u6863\u672c\u8eab\u7684\u5c42\u7ea7\u3002\u804a\u5929\u8bb0\u5f55\u3001\u96f6\u6563\u7b14\u8bb0\u3001\u51e0\u767e\u5b57\u7684 FAQ\uff0c\u6ca1\u6709\u76ee\u5f55\u53ef\u8d70\uff0c\u6811\u4f1a\u9000\u5316\u6210\u4e00\u5806\u7b49\u957f\u7684\u5757\uff0c\u90a3\u65f6\u5019\u5b83\u8ddf\u5411\u91cf\u68c0\u7d22\u6bd4\u6ca1\u6709\u4efb\u4f55\u4f18\u52bf\uff0c\u53cd\u800c\u66f4\u8d35\u3002issue #518 \u5c31\u662f\u8fd9\u4e2a\u95ee\u9898\u7684\u6781\u7aef\u7248\u672c\uff1a\u6709\u4eba\u62ff flash \u6a21\u5f0f\u8dd1 FinanceBench\uff0c\u53d1\u73b0\u4e0d\u5c11 PDF \u7684\u524d\u51e0\u9875\u6839\u672c\u6ca1\u88ab\u62bd\u8fdb\u7d22\u5f15\uff0c\u8986\u76d6\u7387\u53ea\u6709\u4e00\u534a\u3002\u7d22\u5f15\u91cc\u6ca1\u6709\u7684\u9875\uff0c\u540e\u9762\u63a8\u7406\u518d\u806a\u660e\u4e5f\u627e\u4e0d\u5230\u3002<\/p>\n<p><strong>\u591a\u6587\u6863\u4e0d\u884c\uff0c\u81f3\u5c11\u4e0d\u80fd\u76f4\u63a5\u4e0a\u3002<\/strong> \u5f00\u7bb1\u7684\u72b6\u6001\u662f\u4e00\u68f5\u6811\u5bf9\u5e94\u4e00\u4efd\u6587\u6863\u3002\u5168\u90e8\u8bed\u6599\u4e0a\u505a\u68c0\u7d22\uff0c\u8981\u4e48\u81ea\u5df1\u5199\u4e00\u5c42\u8def\u7531\uff0c\u8981\u4e48\u7528\u4ed6\u4eec\u4e91\u7248\u7684\u6587\u4ef6\u7cfb\u7edf\u5c42\u3002\u8fd9\u4e2a\u80fd\u529b\u6211\u6ca1\u770b\u5230\u5f00\u6e90\u7248\u91cc\u6709\u3002<\/p>\n<p><strong>\u8981\u6beb\u79d2\u7ea7\u9996\u5b57\u7684\u4e0d\u884c\u3002<\/strong> \u8fd9\u4e2a\u4e0d\u7528\u89e3\u91ca\u3002<\/p>\n<p>\u8fd8\u6709\u4e00\u4e2a\u6211\u4e2a\u4eba\u7684\u4fdd\u7559\u610f\u89c1\uff1a\u5b83\u628a\u4e0d\u786e\u5b9a\u6027\u4ece\u300c\u5d4c\u5165\u6a21\u578b\u9009\u5f97\u5bf9\u4e0d\u5bf9\u300d\u632a\u5230\u4e86\u300c\u6a21\u578b\u8fd9\u4e00\u6b21\u8d70\u7684\u8def\u5bf9\u4e0d\u5bf9\u300d\u3002\u9519\u8bef\u6ca1\u6d88\u5931\uff0c\u53ea\u662f\u6362\u4e86\u4e2a\u5f62\u72b6\u2014\u2014\u800c\u4e14\u65b0\u5f62\u72b6\u6bd4\u65e7\u5f62\u72b6\u66f4\u96be\u6d4b\u3002\u5411\u91cf\u68c0\u7d22\u53ec\u56de\u9519\u4e86\uff0c\u4f60\u80fd\u770b\u7740\u76f8\u4f3c\u5ea6\u5206\u6570\u9a82\u5b83\uff1b\u6811\u641c\u7d22\u8d70\u9519\u4e86\uff0c\u4f60\u5f97\u628a\u6574\u6761\u63a8\u7406\u8def\u5f84\u7ffb\u51fa\u6765\u770b\uff0c\u624d\u77e5\u9053\u5b83\u662f\u4ece\u54ea\u4e00\u6b65\u5f00\u59cb\u8dd1\u504f\u7684\u3002<\/p>\n<h2 id=\"_3\">\u88c5\u5b83\u4e24\u5206\u949f\uff0c\u627e\u5165\u53e3\u82b1\u4e86\u5341\u5206\u949f<\/h2>\n<pre><code class=\"language-bash\">python3 -m venv \/tmp\/pi\n\/tmp\/pi\/bin\/pip install -U pageindex\n<\/code><\/pre>\n<p>\u88c5\u5b8c\u662f <code>0.4.7<\/code>\uff0cPyPI \u4e0a 10 \u6708 24 \u65e5\u53d1\u7684\u7248\u3002Python \u8981\u6c42 <code>&gt;=3.10<\/code>\uff0c\u6211\u8fd9\u53f0\u662f 3.13.12\uff0c\u6ca1\u95ee\u9898\u3002<\/p>\n<p>\u7b2c\u4e00\u4e2a\u610f\u5916\uff1a\u88c5\u5b8c\u6ca1\u6709 <code>pageindex<\/code> \u8fd9\u4e2a\u547d\u4ee4\u3002<code>pip<\/code> \u53ea\u88c5\u4e86\u5e93\uff0c\u547d\u4ee4\u884c\u811a\u672c <code>run_pageindex.py<\/code> \u8fd8\u8eba\u5728\u4ed3\u5e93\u6839\u76ee\u5f55\uff0c\u5f97\u81ea\u5df1 clone \u4e0b\u6765\u6216\u8005\u5355\u72ec\u4e0b\u8fd9\u4e2a\u6587\u4ef6\u3002\u6587\u6863\u7684\u5feb\u901f\u4e0a\u624b\u5199\u7684\u662f <code>pip install<\/code> \u7136\u540e\u5728 Python \u91cc\u8c03 <code>PageIndexClient<\/code>\uff0c\u811a\u672c\u7528\u6cd5\u5199\u5728\u4ed3\u5e93 README \u91cc\uff0c\u4e24\u5904\u53e3\u5f84\u4e0d\u4e00\u6837\uff0c\u6211\u7b2c\u4e00\u6b21\u5c31\u627e\u9519\u5730\u65b9\u4e86\u3002<\/p>\n<p>\u7b2c\u4e8c\u4e2a\u610f\u5916\u662f\u4f53\u79ef\u3002<code>litellm<\/code>\u3001<code>pypdfium2<\/code>\u3001<code>PyPDF2<\/code>\u3001<code>Pillow<\/code>\uff0c\u518d\u52a0\u4e0a\u4e00\u4e32\u8ddf\u5de5\u5177\u8c03\u7528\u548c\u534f\u8bae\u76f8\u5173\u7684\u4f9d\u8d56\uff0c\u4e00\u8def\u88c5\u4e0b\u6765\u865a\u62df\u73af\u5883\u91cc 81 \u4e2a\u5305\uff0c<code>site-packages<\/code> \u5360 339 MB\u3002\u5bf9\u4e00\u4e2a\u300c\u5c31\u662f\u4e2a\u6587\u6863\u7d22\u5f15\u5668\u300d\u7684\u4e1c\u897f\u6765\u8bf4\u4e0d\u7b97\u5c0f\uff0c\u4f46\u90fd\u5728\u4f9d\u8d56\u6811\u91cc\uff0c\u6ca1\u6709\u5077\u5077\u4e0b\u6a21\u578b\u6743\u91cd\u3002<code>litellm<\/code> \u8fd8\u88ab\u9489\u5728 <code>1.105.x<\/code>\uff08<code>&lt;1.106<\/code>\uff09\uff0c\u8bf4\u660e\u5b83\u8ddf\u7740\u8fd9\u6761\u7ebf\u8d70\uff0c\u5347\u7ea7\u8282\u594f\u5f97\u770b\u4e0a\u6e38\u8138\u8272\u3002<\/p>\n<p>\u786c\u4ef6\u95e8\u69db\u5b98\u65b9\u6ca1\u5199\uff0cREADME \u548c\u6587\u6863\u7ad9\u90fd\u6ca1\u7ed9\u6700\u4f4e\u914d\u7f6e\u3002\u8fd9\u5757\u6211\u53ea\u80fd\u62a5\u6211\u81ea\u5df1\u91cf\u5230\u7684\u6570\uff1a<\/p>\n<table>\n<thead>\n<tr>\n<th>\u6863\u4f4d<\/th>\n<th>\u914d\u7f6e<\/th>\n<th>\u4f9d\u636e<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u6700\u4f4e\u53ef\u7528<\/td>\n<td>2 \u6838 4 GB\uff0c\u7eaf CPU\uff0c20 GB \u78c1\u76d8<\/td>\n<td>\u6211\u5b9e\u6d4b\u4e00\u4efd 12 \u9875 PDF \u79bb\u7ebf\u5efa\u6811\uff1a1.65 \u79d2\uff0c\u5cf0\u503c\u5185\u5b58 148 MB<\/td>\n<\/tr>\n<tr>\n<td>\u5efa\u8bae\u914d\u7f6e<\/td>\n<td>4 \u6838 8 GB \u4ee5\u4e0a<\/td>\n<td>\u6458\u8981\u5e76\u53d1\u9ed8\u8ba4 64 \u8def\uff08<code>SUMMARY_CONCURRENCY = 64<\/code>\uff09\uff0c\u6811\u4f18\u5316\u5c55\u5f00\u9ed8\u8ba4 32 \u8def\uff0c\u5e76\u53d1\u6253\u6ee1\u65f6\u5185\u5b58\u548c\u4f9b\u5e94\u5546\u9650\u6d41\u90fd\u4f1a\u96be\u53d7<\/td>\n<\/tr>\n<tr>\n<td>GPU<\/td>\n<td>\u4e0d\u9700\u8981<\/td>\n<td>\u63a8\u7406\u5168\u5728\u8fdc\u7aef\u6a21\u578b\uff0c\u672c\u5730\u53ea\u505a\u7248\u9762\u89e3\u6790<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u90a3\u4e2a 148 MB \u662f <code>--no-summary<\/code> \u7684\u88f8\u5efa\u6811\uff1b\u5f00\u7740\u6458\u8981\u8dd1\uff0c\u5185\u5b58\u4f1a\u968f\u5e76\u53d1\u4e0a\u53bb\uff0c\u5b98\u65b9\u6ca1\u7ed9\u6570\u5b57\uff0c\u6211\u4e5f\u6ca1\u6709 key \u53bb\u91cf\uff0c\u53ea\u80fd\u8bf4\u5230\u8fd9\u513f\u3002<\/p>\n<h2 id=\"29\">\u4e00\u4efd\u8fea\u58eb\u5c3c\u5b63\u62a5\uff0c29 \u4e2a\u8282\u70b9<\/h2>\n<p>\u7aef\u5230\u7aef\u6211\u53ea\u8dd1\u901a\u4e86\u524d\u534a\u6bb5\uff0c\u540e\u534a\u6bb5\u7f3a key\u3002\u524d\u534a\u6bb5\u7684\u8f93\u5165\u8f93\u51fa\u90fd\u662f\u771f\u7684\u3002<\/p>\n<p>\u8f93\u5165\u7528\u4ed3\u5e93\u81ea\u5e26\u7684\u6837\u4f8b\uff0c\u8fea\u58eb\u5c3c 2025 \u8d22\u5e74\u4e00\u5b63\u5ea6\u8d22\u62a5\uff0c12 \u9875\uff1a<\/p>\n<pre><code class=\"language-bash\">python run_pageindex.py --pdf_path q1.pdf --no-summary --optimize off\n<\/code><\/pre>\n<p>\u9884\u671f\u8f93\u51fa\u5c31\u4e24\u884c\uff1a<\/p>\n<pre><code>Parsing done, saving to file...\nTree structure saved to: .\/results\/q1_structure.json\n<\/code><\/pre>\n<p>\u51fa\u6765\u8fd9\u4e2a\u6587\u4ef6 4,572 \u5b57\u8282\uff0c29 \u4e2a\u8282\u70b9\uff0c\u9876\u5c42 12 \u4e2a\uff0c<code>doc_title<\/code> \u662f\u4ece\u7248\u9762\u91cc\u62bd\u51fa\u6765\u7684\uff0c\u4e00\u4e2a\u5b57\u6ca1\u5dee\uff1a<\/p>\n<pre><code class=\"language-json\">{&quot;doc_name&quot;: &quot;q1.pdf&quot;,\n &quot;doc_title&quot;: &quot;THE WALT DISNEY COMPANY REPORTS FIRST QUARTER EARNINGS FOR FISCAL 2025&quot;,\n &quot;structure&quot;: [\n   {&quot;title&quot;: &quot;Financial Results for the Quarter:&quot;, &quot;node_id&quot;: &quot;0000&quot;, &quot;start_index&quot;: 1, &quot;end_index&quot;: 1},\n   {&quot;title&quot;: &quot;DISCUSSION OF FIRST QUARTER SEGMENT RESULTS&quot;, &quot;node_id&quot;: &quot;0006&quot;, &quot;start_index&quot;: 4, &quot;end_index&quot;: 9,\n    &quot;nodes&quot;: [\n      {&quot;title&quot;: &quot;Star India&quot;, &quot;node_id&quot;: &quot;0007&quot;, &quot;start_index&quot;: 4, &quot;end_index&quot;: 4},\n      {&quot;title&quot;: &quot;Entertainment&quot;, &quot;node_id&quot;: &quot;0008&quot;, &quot;start_index&quot;: 4, &quot;end_index&quot;: 7}\n    ]}]}\n<\/code><\/pre>\n<p><code>start_index<\/code> \u548c <code>end_index<\/code> \u662f\u9875\u7801\u533a\u95f4\u3002\u770b\u5230\u8fd9\u4e2a\u7ed3\u6784\u6211\u5c31\u660e\u767d\u5b83\u4e3a\u4ec0\u4e48\u6562\u8bf4\u300c\u4e0d\u505a chunking\u300d\u4e86\u2014\u2014\u5b83\u786e\u5b9e\u4e0d\u6309\u5b57\u6570\u5207\uff0c\u5b83\u6309\u7ae0\u8282\u5207\u3002\u53ea\u662f\u8fd9\u53e5\u8bdd\u7684\u53e6\u4e00\u9762\u662f\uff1a<strong>\u5b83\u6362\u4e86\u4e2a\u5730\u65b9\u5207<\/strong>\u3002\u7ae0\u8282\u5207\u5f97\u597d\u4e0d\u597d\uff0c\u53d6\u51b3\u4e8e\u4f60\u7684 PDF \u6709\u6ca1\u6709\u50cf\u6837\u7684\u6807\u9898\u5c42\u7ea7\u3002<\/p>\n<p>\u628a <code>--optimize off<\/code> \u6362\u6210 <code>--optimize full<\/code>\uff0c\u591a\u6253\u4e00\u884c\uff1a<\/p>\n<pre><code>Optimize: merges=1 expands=0, worst-case search cost 6 -&gt; 6 pages\n<\/code><\/pre>\n<p>\u7ffb\u8bd1\u6210\u4eba\u8bdd\uff1a\u5b83\u628a\u4e00\u4e2a\u5c0f\u8282\u70b9\u5408\u8fdb\u4e86\u7236\u8282\u70b9\uff0c\u7136\u540e\u7b97\u51fa\u6700\u574f\u60c5\u51b5\u4e0b\u4e00\u6b21\u641c\u7d22\u8981\u7ffb 6 \u9875\u3002\u8fd9\u4e2a\u300cworst-case search cost\u300d\u662f\u6211\u5728\u8fd9\u5957\u4e1c\u897f\u91cc\u770b\u5230\u7684\u6700\u8bda\u5b9e\u7684\u4e00\u4e2a\u6307\u6807\u2014\u2014\u5b83\u81ea\u5df1\u627f\u8ba4\uff0c\u4ece\u6839\u8d70\u5230\u53f6\u5b50\uff0c\u6700\u574f\u8981\u4ed8\u51fa\u591a\u5c11\u9875\u7684\u9605\u8bfb\u91cf\u3002<\/p>\n<p>\u518d\u5f80\u4e0b\u5c31\u8d70\u4e0d\u52a8\u4e86\u3002\u53bb\u6389 <code>--no-summary<\/code>\u3001\u7528\u9ed8\u8ba4\u7684 flash \u6a21\u5f0f\uff0c\u65e5\u5fd7\u5f00\u59cb\u5237\u540c\u4e00\u6761\u9519\u8bef\uff0c\u7b2c\u5341\u6b21\u4e4b\u540e\u629b\u51fa\u6765\uff1a<\/p>\n<pre><code>pageindex.utils.LLMRetriesExhausted: LLM completion failed after 10 retries:\nInternalServerError ... Missing credentials\n<\/code><\/pre>\n<p>SDK \u90a3\u8fb9\u540c\u6837\u5361\u4f4f\uff0c\u629b\u7684\u662f <code>PageIndexAPIError<\/code>\uff0c12.3 \u79d2\u4e4b\u540e\u653e\u5f03\uff0c<code>storage_path<\/code> \u9ed8\u8ba4\u843d\u5728\u5f53\u524d\u76ee\u5f55\u7684 <code>.pageindex<\/code>\u3002<\/p>\n<p>\u6211\u8fd9\u513f\u6ca1\u6709\u53ef\u7528\u7684\u6a21\u578b key\uff0c\u4e5f\u6ca1\u6709\u628a\u516c\u53f8\u5408\u540c\u5f80\u5916\u90e8\u63a5\u53e3\u4e0a\u53d1\u7684\u6253\u7b97\uff0c\u6240\u4ee5\u6458\u8981\u548c\u68c0\u7d22\u8fd9\u4e24\u6b65\u6211\u6ca1\u8dd1\u3002\u90a3\u4e2a 98.7% \u6211\u9a8c\u4e0d\u4e86\uff0c\u4e5f\u4e0d\u66ff\u5b83\u80cc\u4e66\u3002\u80fd\u66ff\u5b83\u80cc\u4e66\u7684\u53ea\u6709\u4e00\u4ef6\u4e8b\uff1a<strong>\u79bb\u7ebf\u5efa\u6811\u8fd9\u4e00\u6b65\u662f\u771f\u7684\uff0c\u800c\u4e14\u4e0d\u9700\u8981\u4efb\u4f55\u5916\u90e8\u670d\u52a1<\/strong>\u3002<\/p>\n<h2 id=\"issue\">\u7ffb issue \u7ffb\u51fa\u6765\u7684\u4e1c\u897f\uff0c\u6bd4\u6587\u6863\u91cc\u5199\u5f97\u591a<\/h2>\n<p>\u6309\u8bf4\u8fd9\u79cd\u4e1c\u897f\u7684\u5751\u90fd\u85cf\u5728 issue \u91cc\uff0c\u6211\u7ffb\u4e86\u4e00\u904d\uff0c\u6309\u4efd\u91cf\u6392\uff1a<\/p>\n<p><strong>#518\uff0c\u8986\u76d6\u7387\u53ea\u6709\u4e00\u534a\u3002<\/strong> \u4e00\u4f4d\u7528\u6237\u5728 FinanceBench \u6570\u636e\u96c6\u4e0a\u8dd1 flash \u6a21\u5f0f\uff0c\u53d1\u73b0\u4e0d\u5c11 PDF \u7684\u524d\u51e0\u9875\u6ca1\u88ab\u62bd\u51fa\u6765\uff0c\u6bd4\u5982 <code>3M_2108k.pdf<\/code>\uff0c\u7d22\u5f15\u8986\u76d6\u7387\u62a5\u7684\u662f 50%\u3002\u8fd9\u6761\u5230\u73b0\u5728\u8fd8\u662f open\u3002\u5b83\u6233\u7684\u6b63\u662f\u6574\u4e2a\u65b9\u6848\u7684\u5730\u57fa\u2014\u2014\u7d22\u5f15\u91cc\u6ca1\u6709\u7684\u9875\uff0c\u63a8\u7406\u518d\u51c6\u4e5f\u6ca1\u7528\u3002<\/p>\n<p><strong>#316\uff0c\u7d22\u5f15\u53ea\u80fd\u6574\u4f53\u91cd\u5efa\uff088 \u6761\u8ba8\u8bba\uff09\u3002<\/strong> \u4e00\u4efd 500 \u9875\u7684\u6587\u6863\u6539\u4e86\u4e00\u8282\uff0c\u73b0\u5728\u5f97\u6574\u4efd\u91cd\u8dd1\u4e00\u904d\uff0c\u5efa\u6811\u6210\u672c\u548c token \u6210\u672c\u4e00\u8d77\u91cd\u4ed8\u3002\u63d0\u9700\u6c42\u7684\u4eba\u8981\u7684\u662f\u589e\u91cf\u66f4\u65b0\uff0c\u5b98\u65b9\u8fd8\u6ca1\u7ed9\u3002\u8fd9\u4e2a\u5751\u5bf9\u9759\u6001\u6587\u6863\u65e0\u6240\u8c13\uff0c\u5bf9\u7ecf\u5e38\u6539\u7248\u7684\u5408\u540c\u6a21\u677f\u5c31\u5f88\u75bc\u3002<\/p>\n<p><strong>#268\uff0c\u8f93\u51fa\u8def\u5f84\u5199\u6b7b\u3002<\/strong> <code>.\/results\/{name}_structure.json<\/code> \u662f\u786c\u7f16\u7801\u7684\uff0c\u60f3\u6279\u91cf\u5904\u7406\u5c31\u53ea\u80fd\u4e8b\u540e\u642c\u6587\u4ef6\u3002\u5199\u811a\u672c\u7684\u4eba\u4f1a\u5148\u649e\u4e0a\u8fd9\u4e2a\u3002<\/p>\n<p><strong>#456\uff0cWindows \u4e0a\u7684\u9501\u662f\u5047\u7684\u3002<\/strong> <code>DocStore.lock()<\/code> \u5728 Windows \u4e0a\u9000\u5316\u6210\u4e00\u4e2a\u4ec0\u4e48\u90fd\u4e0d\u505a\u7684 <code>yield<\/code>\uff0c\u56e0\u4e3a <code>fcntl.flock<\/code> \u5728\u90a3\u8fb9\u4e0d\u5b58\u5728\u3002\u5e76\u53d1\u63d0\u4ea4\u540c\u540d\u6587\u6863\u65f6\uff0c\u90a3\u4e2a\u300c\u5148\u67e5\u91cd\u518d\u5199\u300d\u7684\u7a97\u53e3\u662f\u655e\u5f00\u7684\u3002<\/p>\n<p><strong>#103\uff0c\u88c5\u4e0d\u4e0a\u8fc7\u3002<\/strong> \u65e9\u671f\u4ed3\u5e93\u91cc\u6ca1\u6709 <code>pyproject.toml<\/code>\uff0c<code>pip install<\/code> \u76f4\u63a5\u5931\u8d25\u3002\u8fd9\u6761\u5df2\u7ecf\u5173\u4e86\uff0c\u73b0\u5728\u88c5\u5f97\u4e0a\u2014\u2014\u4f46\u7ffb issue \u5217\u8868\u80fd\u770b\u51fa\u8fd9\u4e2a\u9879\u76ee\u7684\u5de5\u7a0b\u5316\u662f\u540e\u6765\u8865\u7684\u3002<\/p>\n<p>\u5916\u90e8\u7684\u8ba8\u8bba\u6211\u4e5f\u7ffb\u4e86\u51e0\u5904\u3002Hacker News \u4e0a\u90a3\u6b21 Show HN \u662f 2025 \u5e74 8 \u6708 27 \u65e5\u53d1\u7684\uff0c192 \u5206\u3001128 \u6761\u8bc4\u8bba\uff0c\u6807\u9898\u5f88\u76f4\u767d\uff1aShow HN: PageIndex \u2013 Vectorless RAG\u3002\u8bc4\u8bba\u533a\u6700\u54cd\u7684\u4e00\u53e5\u8d28\u7591\u662f\u300c\u8fd9\u4e0d\u5c31\u662f\u6362\u4e86\u4e2a\u5730\u65b9\u7684 vibe retrieval \u5417\u300d\u2014\u2014\u539f\u6765\u9760\u5411\u91cf\u7b97\u8fd1\u4f3c\uff0c\u73b0\u5728\u9760\u6a21\u578b\u731c\u8def\u5f84\uff0c\u90fd\u662f\u8fd1\u4f3c\uff0c\u51ed\u4ec0\u4e48\u8bf4\u540e\u8005\u66f4\u51c6\u3002\u8fd9\u8bdd\u6211\u8ba4\u540c\u4e00\u534a\uff1a\u5b83\u786e\u5b9e\u628a\u786e\u5b9a\u6027\u8ba1\u7b97\u6362\u6210\u4e86\u968f\u673a\u8c03\u7528\uff0c\u53ea\u4e0d\u8fc7\u6362\u6765\u7684\u4e1c\u897f\u662f\u300c\u53ef\u89e3\u91ca\u300d\uff0c\u800c\u4e0d\u662f\u300c\u66f4\u51c6\u300d\u3002<\/p>\n<p>\u8fd8\u6709\u4eba\u8d34\u4e86\u5ef6\u8fdf\u7684\u5b9e\u6d4b\uff0c\u5355\u6b21\u67e5\u8be2\u51e0\u5341\u79d2\u5230\u4e24\u5206\u949f\uff0c\u53d6\u51b3\u4e8e\u6587\u6863\u957f\u5ea6\u548c\u6a21\u578b\uff1b\u4e5f\u6709\u4eba\u6307\u51fa\u5b83\u672c\u8d28\u4e0a\u662f\u300c\u62ff\u6a21\u578b\u7ed9\u6bcf\u4e00\u8282\u5199\u6458\u8981\uff0c\u518d\u628a\u6811\u585e\u8fdb\u4e0a\u4e0b\u6587\u8ba9\u6a21\u578b\u6307\u8282\u70b9\u300d\uff0c\u4e00\u4f4d\u81ea\u5df1\u5b9e\u73b0\u7b80\u5316\u7248\u7684\u7528\u6237\u8bf4\u7528\u66f4\u5c11\u7684\u8c03\u7528\u5728 FinanceBench \u4e0a\u62ff\u5230\u8fc7\u516b\u6210\u51fa\u5934\u3002<\/p>\n<p>\u4e2d\u6587\u5708\u7684\u5b9e\u6d4b\u63d0\u5230\u4e24\u4e2a\u5177\u4f53\u7684\u5751\uff1a\u4e00\u662f\u6ca1\u6709\u6807\u9898\u7684\u6587\u6863\uff08\u804a\u5929\u8bb0\u5f55\u3001\u96f6\u6563\u7b14\u8bb0\uff09\u6839\u672c\u751f\u6210\u4e0d\u4e86\u50cf\u6837\u7684\u6811\uff1b\u4e8c\u662f\u8d70\u4e91\u7248 API \u65f6 <code>submit_document()<\/code> \u4f1a\u628a\u6574\u4efd PDF \u4f20\u4e0a\u53bb\uff0c\u516c\u5f00\u8bba\u6587\u65e0\u6240\u8c13\uff0c\u5408\u540c\u3001\u8d22\u62a5\u8fd9\u79cd\u5c31\u5f97\u5148\u770b\u6570\u636e\u653f\u7b56\u3002<\/p>\n<h2 id=\"_4\">\u8c01\u8be5\u4e0a\uff0c\u8c01\u522b\u4e0a<\/h2>\n<p>\u6211\u7684\u7ed3\u8bba\u5206\u4e09\u5c42\u3002<\/p>\n<p><strong>\u9002\u5408\u4e0a\u7684<\/strong>\uff1a\u8bed\u6599\u662f\u957f\u3001\u7ed3\u6784\u5316\u3001\u4e13\u4e1a\u6587\u6863\uff08\u8d22\u62a5\u3001\u62db\u80a1\u4e66\u3001\u5408\u540c\u3001\u6cd5\u89c4\u3001\u6280\u672f\u624b\u518c\uff09\uff1b\u67e5\u8be2\u4f4e\u9891\u4f46\u5355\u6b21\u4ef7\u503c\u9ad8\uff1b\u5bf9\u6eaf\u6e90\u6709\u786c\u8981\u6c42\uff0c\u51fa\u4e86\u7b54\u6848\u8981\u80fd\u6307\u5230\u7b2c\u51e0\u9875\u7b2c\u51e0\u8282\uff1b\u6216\u8005\u4f60\u5df2\u7ecf\u628a\u6df7\u68c0\u52a0\u91cd\u6392\u8c03\u5230\u4e86\u74f6\u9888\uff0c\u957f\u6587\u6863\u7684\u591a\u8df3\u95ee\u9898\u8fd8\u662f\u7b54\u4e0d\u51c6\u3002\u8fd9\u51e0\u79cd\u60c5\u51b5\u5b83\u7ed9\u4f60\u7684\u4e1c\u897f\uff0c\u5411\u91cf\u68c0\u7d22\u786e\u5b9e\u7ed9\u4e0d\u4e86\u3002<\/p>\n<p><strong>\u522b\u4e0a\u7684<\/strong>\uff1a\u6d77\u91cf\u788e\u7247\u8bed\u6599\uff1b\u9ad8\u5e76\u53d1\u5728\u7ebf\u95ee\u7b54\uff1b\u9884\u7b97\u6309 QPS \u7b97\u800c\u4e0d\u662f\u6309\u6b21\u7b97\uff1b\u6587\u6863\u5929\u5929\u6539\u7248\u3002\u8fd9\u56db\u79cd\u60c5\u51b5\u786c\u4e0a\uff0c\u4f60\u4f1a\u5f97\u5230\u4e00\u4e2a\u53c8\u6162\u53c8\u8d35\u7684\u5411\u91cf\u5e93\u66ff\u4ee3\u54c1\u3002<\/p>\n<p><strong>\u66ff\u4ee3\u65b9\u6848<\/strong>\uff0c\u6211\u62ff\u4e24\u4e2a\u6bd4\uff1a<\/p>\n<p><strong>RAGFlow<\/strong>\uff08<code>infiniflow\/ragflow<\/code>\uff0c92,919 star\uff0cApache-2.0\uff09\u3002\u8d70\u7684\u662f\u5b8c\u5168\u76f8\u53cd\u7684\u8def\u7ebf\u2014\u2014\u628a\u89e3\u6790\u3001\u5207\u5757\u3001\u5411\u91cf\u3001\u91cd\u6392\u3001\u5bf9\u8bdd\u5168\u94fe\u8def\u6253\u5305\u6210\u4e00\u5957 Docker Compose\uff0c\u5de5\u7a0b\u5b8c\u6574\u5ea6\u662f\u5b83\u7684\u5f3a\u9879\u3002\u7ef4\u5ea6\u5dee\u522b\u5728<strong>\u4f60\u8981\u4e0d\u8981\u81ea\u5df1\u9020<\/strong>\uff1a\u60f3\u5f00\u7bb1\u6709\u4e2a\u80fd\u7528\u7684\u77e5\u8bc6\u5e93\u7cfb\u7edf\uff0c\u9009\u5b83\uff1b\u60f3\u5728\u81ea\u5df1\u4ee3\u7801\u91cc\u6362\u6389\u68c0\u7d22\u8fd9\u4e00\u5c42\uff0cPageIndex \u66f4\u8f7b\uff0c\u4e00\u4e2a pip \u5305\uff0c\u6ca1\u6709\u670d\u52a1\u8981\u8fd0\u7ef4\u3002<\/p>\n<p><strong>Docling<\/strong>\uff08<code>docling-project\/docling<\/code>\uff0c69,640 star\uff0cMIT\uff09\u3002\u5b83\u53ea\u505a\u524d\u534a\u6bb5\u2014\u2014\u628a PDF \u53d8\u6210\u7ed3\u6784\u5316 Markdown\uff0c\u6807\u9898\u5c42\u7ea7\u3001\u8868\u683c\u3001\u9605\u8bfb\u987a\u5e8f\u90fd\u7ed9\u4f60\u4fdd\u4f4f\u3002\u7ef4\u5ea6\u5dee\u522b\u5728<strong>\u8fb9\u754c<\/strong>\uff1aDocling \u5230\u6587\u6863\u7ed3\u6784\u4e3a\u6b62\uff0c\u68c0\u7d22\u5f97\u4f60\u81ea\u5df1\u63a5\uff1bPageIndex \u628a\u7ed3\u6784\u548c\u68c0\u7d22\u4e00\u8d77\u7ed9\u4e86\uff0c\u4f46\u7ed3\u6784\u90a3\u4e00\u6bb5\u6ca1\u6709 Docling \u7ec6\u3002\u6211\u73b0\u5728\u7684\u60f3\u6cd5\u662f\u8fd9\u4fe9\u53ef\u4ee5\u53e0\u7740\u7528\uff0c\u89e3\u6790\u7528 Docling\uff0c\u68c0\u7d22\u8fd9\u4e00\u5c42\u518d\u51b3\u5b9a\u7528\u6811\u8fd8\u662f\u7528\u5411\u91cf\u3002<\/p>\n<p>\u987a\u5e26\u63d0\u4e00\u53e5 MinerU\uff0882,538 star\uff09\uff0c\u5b9a\u4f4d\u8ddf Docling \u5f88\u8fd1\uff0c\u4e2d\u6587 PDF \u7684\u7248\u9762\u8fd8\u539f\u505a\u5f97\u4e0d\u9519\uff0c\u4e5f\u662f\u53ea\u505a\u89e3\u6790\u8fd9\u4e00\u6bb5\u3002<\/p>\n<p>\u4e00\u53e5\u8bdd\uff1a<strong>PageIndex \u6cbb\u7684\u662f\u300c\u76f8\u4f3c\u4f46\u4e0d\u662f\u8981\u627e\u7684\u90a3\u4e00\u6bb5\u300d\uff0c\u4e0d\u6cbb\u300c\u6a21\u578b\u8bfb\u4e0d\u61c2\u90a3\u4e00\u6bb5\u300d\u3002<\/strong> \u524d\u8005\u662f\u6211\u5929\u5929\u9047\u5230\u7684\uff0c\u540e\u8005\u5f97\u9760\u522b\u7684\u529e\u6cd5\u3002<\/p>\n<p>\u603b\u800c\u8a00\u4e4b\uff0c\u8fd9\u4e1c\u897f\u7684\u65b9\u5411\u6211\u8ba4\uff0c\u5de5\u7a0b\u6210\u719f\u5ea6\u6211\u6253\u4e2a\u6298\u3002\u987a\u624b\u770b\u4e86\u4e0b\u4ed3\u5e93\uff0c\u73b0\u5728 39,794 \u4e2a star\u30013,432 \u4e2a fork\uff0c\u6700\u8fd1\u56db\u5468 77 \u6b21\u63d0\u4ea4\uff0c\u6700\u65b0\u53d1\u7248 <code>v0.4.7<\/code> \u662f 10 \u6708 24 \u65e5\uff0crelease note \u7684\u6807\u9898\u5f88\u81ea\u4fe1\u2014\u2014&#8221;Better retrieval than Vector RAG, at lower cost&#8221;\u3002<\/p>\n<p>\u6700\u540e\u8bf4\u53e5\u5b9e\u5728\u7684\uff0c\u6211\u8fd9\u6b21\u7684\u7ed3\u8bba\u53ea\u8d70\u4e86\u4e00\u534a\u8def\uff1a\u6811\u5efa\u51fa\u6765\u4e86\uff0c\u68c0\u7d22\u90a3\u4e00\u6bb5\u56e0\u4e3a\u6ca1 key \u505c\u5728\u90a3\u513f\u3002\u7b49\u54ea\u5929\u624b\u4e0a\u6709\u5408\u9002\u7684\u6a21\u578b\u989d\u5ea6\u3001\u53c8\u62ff\u5230\u4e00\u4efd\u53ef\u4ee5\u5916\u53d1\u7684\u516c\u5f00\u8d22\u62a5\uff0c\u6211\u4f1a\u628a\u5b83\u548c\u73b0\u5728\u8fd9\u5957\u6df7\u68c0\u6446\u5728\u4e00\u8d77\uff0c\u7528\u540c\u4e00\u6279\u95ee\u9898\u8dd1\u4e00\u904d\uff0c\u770b\u8c01\u7b54\u5f97\u51c6\u3002\u5728\u90a3\u4e4b\u524d\uff0c\u5b83\u5bf9\u6211\u6765\u8bf4\u662f\u4e00\u4e2a\u65b9\u5411\u5bf9\u3001\u5730\u57fa\u8fd8\u6ca1\u9a8c\u6536\u5b8c\u7684\u4e1c\u897f\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5408\u540c\u5207\u516b\u767e\u5b57\u4e00\u5757\uff0c\u95ee\u8fdd\u7ea6\u91d1\u53ec\u56de\u516b\u6bb5\u5168\u662f\u4ed8\u6b3e\u6761\u6b3e\u3002\u6211\u6362\u4e86\u4e2a\u601d\u8def\uff1aPageIndex \u628a\u6587\u6863\u76ee\u5f55\u505a\u6210\u6811\u8ba9\u6a21\u578b\u8d70\u3002\u79bb\u7ebf\u5efa\u6811\u8dd1\u901a\u4e86\uff0c\u68c0\u7d22\u5361\u5728\u6ca1\u6709 key\u300239,794 star\uff0c\u65b9\u5411\u6211\u8ba4\uff0c\u5730\u57fa\u8fd8\u6ca1\u9a8c\u6536\u5b8c\u3002<\/p>\n","protected":false},"author":1,"featured_media":10021507,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[4],"tags":[],"class_list":["post-10021508","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-share"],"jetpack_sharing_enabled":true,"jetpack_featured_media_url":"https:\/\/itmax.cn\/wp-content\/uploads\/2026\/10\/pageindex-cover.png","_links":{"self":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts\/10021508","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=10021508"}],"version-history":[{"count":1,"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts\/10021508\/revisions"}],"predecessor-version":[{"id":10021509,"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts\/10021508\/revisions\/10021509"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/media\/10021507"}],"wp:attachment":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=10021508"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=10021508"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=10021508"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}