{"id":5773,"date":"2024-09-05T14:08:25","date_gmt":"2024-09-05T06:08:25","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=5773"},"modified":"2024-09-05T14:09:24","modified_gmt":"2024-09-05T06:09:24","slug":"quanqiushougeduoyuchai","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/ja\/quanqiushougeduoyuchai\/","title":{"rendered":"\u5168\u7403\u9996\u4e2a\u591a\u8bed\u8a00 ColBERT\uff1aJina ColBERT V2 \u548c\u5b83\u7684\u2018\u4fc4\u7f57\u65af\u5957\u5a03\u2019\u6280\u672f"},"content":{"rendered":"<p data-tool=\"mdnice\u7f16\u8f91\u5668\"><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter wp-image-5774\" title=\"-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/bd0bb4290de67a0.jpg\" alt=\"-1\" width=\"734\" height=\"385\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/bd0bb4290de67a0.jpg 1080w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/bd0bb4290de67a0-300x158.jpg 300w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/bd0bb4290de67a0-1024x538.jpg 1024w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/bd0bb4290de67a0-768x403.jpg 768w\" sizes=\"auto, (max-width: 734px) 100vw, 734px\" \/><\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u5728\u00a0<span class=\"wx_search_keyword_wrap\"><a href=\"https:\/\/www.kdjingpai.com\/en\/rag\/\">RAG<\/a><i class=\"wx_search_keyword\"><\/i><\/span>\u00a0\u9886\u57df\uff0c\u591a\u5411\u91cf\u6a21\u578b\u00a0ColBERT\u00a0\u901a\u8fc7\u4e3a\u6587\u6863\u7684\u6bcf\u4e2a <a href=\"https:\/\/www.kdjingpai.com\/en\/tokenization\/\">token<\/a> \u751f\u6210\u72ec\u7acb\u7684\u5411\u91cf\uff0c\u5e26\u6765\u4e86\u68c0\u7d22\u7cbe\u5ea6\u7684\u63d0\u5347\u3002\u4f46\u540c\u6837\u4e5f\u5e26\u6765\u4e86\u5b58\u50a8\u9700\u6c42\u7684\u5267\u589e\uff0c\u5e76\u4e14\u4ec5\u652f\u6301\u82f1\u6587\uff0c\u9650\u5236\u4e86\u5176\u5e94\u7528\u8303\u56f4\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u4e3a\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\uff0c\u6211\u4eec\u6539\u8fdb\u4e86 ColBERT \u7684\u67b6\u6784\u548c\u8bad\u7ec3\u6d41\u7a0b\uff0c\u7279\u522b\u662f\u5728\u591a\u8bed\u8a00\u5904\u7406\u65b9\u9762\u53d6\u5f97\u4e86\u7a81\u7834\u3002\u6700\u65b0\u7684 Jina-ColBERT-v2 \u652f\u6301 89 \u79cd\u8bed\u8a00\uff0c\u5e76\u5f15\u5165\u4e86\u81ea\u5b9a\u4e49\u8f93\u51fa\u7ef4\u5ea6\u9009\u9879\uff0c\u663e\u8457\u51cf\u5c11\u5b58\u50a8\u9700\u6c42\uff0c\u63d0\u5347\u4e86\u591a\u8bed\u8a00\u68c0\u7d22\u7684\u6548\u7387\u548c\u51c6\u786e\u6027\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\n<h2 data-tool=\"mdnice\u7f16\u8f91\u5668\">\u65b0\u7248\u672c\u7684\u6838\u5fc3\u4eae\u70b9<\/h2>\n<ol class=\"list-paddingleft-1\" data-tool=\"mdnice\u7f16\u8f91\u5668\">\n<li>\n<section><strong>\u6027\u80fd\u589e\u5f3a<\/strong>\uff1a\u4e0e\u539f\u59cb ColBERT-v2 \u76f8\u6bd4\uff0c\u82f1\u6587\u68c0\u7d22\u6027\u80fd\u63d0\u5347\u4e86 6.5%\uff1b\u76f8\u8f83\u4e8e\u524d\u4ee3 jina-colbert-v1-en\uff0c\u6027\u80fd\u4e5f\u63d0\u5347\u4e86 5.4%\u3002<\/section>\n<\/li>\n<li>\n<section><strong>\u591a\u8bed\u8a00\u652f\u6301<\/strong>\uff1a\u65b0\u7248\u672c\u652f\u6301\u591a\u8fbe 89 \u79cd\u8bed\u8a00\uff0c\u6db5\u76d6\u963f\u62c9\u4f2f\u8bed\u3001\u4e2d\u6587\u3001\u82f1\u8bed\u3001\u65e5\u8bed\u3001\u4fc4\u8bed\u7b49\u591a\u79cd\u8bed\u8a00\uff0c\u540c\u65f6\u652f\u6301\u7f16\u7a0b\u8bed\u8a00\u3002<\/section>\n<\/li>\n<li>\n<section><strong>\u8f93\u51fa\u7ef4\u5ea6\u53ef\u5b9a\u5236<\/strong>\uff1a\u65b0\u7248\u672c\u91c7\u7528\u201c\u4fc4\u7f57\u65af\u5957\u5a03\u201d\u5f0f\u8868\u5f81\u5b66\u4e60\u6280\u672f(Matryoshka Representation Learning, MRL)\uff0c\u63d0\u4f9b 128\u300196 \u548c 64 \u7ef4\u7684\u8f93\u51fa\u5411\u91cf\u9009\u9879\uff0c\u5141\u8bb8\u7528\u6237\u6839\u636e\u5b9e\u9645\u9700\u6c42\u9009\u62e9\u5408\u9002\u7684\u7ef4\u5ea6\u3002<\/section>\n<\/li>\n<\/ol>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u5b8c\u6574\u7684\u6280\u672f\u62a5\u544a\u53ef\u4ee5\u5728 arXiv \u4e0a\u627e\u5230\uff1ahttps:\/\/arxiv.org\/abs\/2408.16672<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\n<h2 data-tool=\"mdnice\u7f16\u8f91\u5668\">jina-colbert-v2 \u8868\u73b0\u4e00\u89c8<\/h2>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u5728\u68c0\u7d22\u6027\u80fd\u4e0a\uff0cjina-colbert-v2 \u5c55\u73b0\u4e86\u4e0d\u4fd7\u7684\u4f18\u52bf\uff0c\u65e0\u8bba\u662f\u82f1\u6587\u68c0\u7d22\u4efb\u52a1\uff0c\u8fd8\u662f\u591a\u8bed\u8a00\u652f\u6301\u90fd\u4f18\u4e8e\u4e4b\u524d\u7684\u7248\u672c\u3002\u901a\u8fc7 Jina AI \u7956\u4f20 8192 token \u8f93\u5165\u957f\u5ea6\u7684\u52a0\u6301\uff0c\u8fd9\u6b3e\u591a\u5411\u91cf\u6a21\u578b\u7684\u4f18\u52bf\u4e5f\u80fd\u5f97\u4ee5\u5145\u5206\u53d1\u6325\u3002\u4ee5\u4e0b\u662f\u4e0e\u5176\u4ed6\u7248\u672c\u7684\u5bf9\u6bd4\uff0c\u53ef\u4ee5\u6e05\u6670\u770b\u5230\u6838\u5fc3\u7684\u6539\u8fdb\uff1a<\/p>\n<section data-tool=\"mdnice\u7f16\u8f91\u5668\">\n<table>\n<thead>\n<tr>\n<th><\/th>\n<th><code>jina-colbert-v2<\/code><\/th>\n<th><code>jina-colbert-v1-en<\/code><\/th>\n<th>\u539f\u59cb ColBERTv2<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u82f1\u8bed BEIR \u4efb\u52a1\u7684\u5e73\u5747\u503c<\/td>\n<td><strong>0.521<\/strong><\/td>\n<td>0.494<\/td>\n<td>0.489<\/td>\n<\/tr>\n<tr>\n<td>\u591a\u8bed\u8a00\u652f\u6301<\/td>\n<td><strong>89 \u79cd\u8bed\u8a00<\/strong><\/td>\n<td>\u4ec5\u652f\u6301\u82f1\u8bed<\/td>\n<td>\u4ec5\u652f\u6301\u82f1\u8bed<\/td>\n<\/tr>\n<tr>\n<td>\u8f93\u51fa\u7ef4\u5ea6<\/td>\n<td><strong>128\u300196 \u6216 64<\/strong><\/td>\n<td>\u56fa\u5b9a 128<\/td>\n<td>\u56fa\u5b9a 128<\/td>\n<\/tr>\n<tr>\n<td>\u6700\u5927\u67e5\u8be2\u957f\u5ea6<\/td>\n<td>32 <a href=\"https:\/\/www.kdjingpai.com\/en\/tokenization\/\">tokens<\/a><\/td>\n<td>32 tokens<\/td>\n<td>32 tokens<\/td>\n<\/tr>\n<tr>\n<td>\u6700\u5927\u6587\u6863\u957f\u5ea6<\/td>\n<td>8192 tokens<\/td>\n<td>8192 tokens<\/td>\n<td>512 tokens<\/td>\n<\/tr>\n<tr>\n<td>\u53c2\u6570\u6570\u91cf<\/td>\n<td>5.6 \u4ebf<\/td>\n<td>1.37 \u4ebf<\/td>\n<td>1.1 \u4ebf<\/td>\n<\/tr>\n<tr>\n<td>\u6a21\u578b\u5927\u5c0f<\/td>\n<td>1.1GB<\/td>\n<td>550MB<\/td>\n<td>438MB<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/section>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\n<h3 data-tool=\"mdnice\u7f16\u8f91\u5668\">1. \u6027\u80fd\u63d0\u5347<\/h3>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u5728\u82f1\u8bed\u68c0\u7d22\u4efb\u52a1\u4e2d\uff0cjina-colbert-v2 \u7684\u8868\u73b0\u8d85\u8d8a\u4e86\u524d\u4e00\u4ee3 jina-colbert-v1-en \u548c\u539f\u7248 ColBERT v2\uff0c\u63a5\u8fd1\u4e13\u4e3a\u82f1\u8bed\u8bbe\u8ba1\u7684 AnswerAI-ColBERT-small \u6a21\u578b\u6c34\u5e73\u3002<\/p>\n<section data-tool=\"mdnice\u7f16\u8f91\u5668\">\n<table>\n<thead>\n<tr>\n<th>\u6a21\u578b\u540d\u79f0<\/th>\n<th>\u82f1\u8bed BEIR \u57fa\u51c6\u6d4b\u8bd5\u7684\u5e73\u5747\u5206<\/th>\n<th>\u591a\u8bed\u8a00\u652f\u6301<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><code>jina-colbert-v2<\/code><\/td>\n<td>0.521<\/td>\n<td><strong>\u652f\u6301\u591a\u8bed\u8a00<\/strong><\/td>\n<\/tr>\n<tr>\n<td><code>jina-colbert-v1-en<\/code><\/td>\n<td>0.494<\/td>\n<td>\u4ec5\u652f\u6301\u82f1\u8bed<\/td>\n<\/tr>\n<tr>\n<td>ColBERT v2.0<\/td>\n<td>0.489<\/td>\n<td>\u4ec5\u652f\u6301\u82f1\u8bed<\/td>\n<\/tr>\n<tr>\n<td>AnswerAI-ColBERT-small<\/td>\n<td>0.549<\/td>\n<td>\u4ec5\u652f\u6301\u82f1\u8bed<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/section>\n<figure data-tool=\"mdnice\u7f16\u8f91\u5668\"><figcaption><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter wp-image-5775\" title=\"-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/406cc5905776546.png\" alt=\"-1\" width=\"770\" height=\"491\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/406cc5905776546.png 1080w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/406cc5905776546-300x191.png 300w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/406cc5905776546-1024x652.png 1024w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/406cc5905776546-768x489.png 768w\" sizes=\"auto, (max-width: 770px) 100vw, 770px\" \/>Jina ColBERT v2 \u5728\u57fa\u4e8e\u82f1\u8bed\u7684 BEIR \u6570\u636e\u96c6\u4e0a\u7684\u8868\u73b0<\/figcaption><\/figure>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\n<h3 data-tool=\"mdnice\u7f16\u8f91\u5668\">2. \u591a\u8bed\u8a00\u652f\u6301<\/h3>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\"><strong>Jina-ColBERT-v2 \u662f\u76ee\u524d\u552f\u4e00\u4e00\u6b3e\u652f\u6301\u591a\u79cd\u8bed\u8a00\u7684 ColBERT \u6a21\u578b<\/strong>\uff0c\u80fd\u591f\u751f\u6210\u975e\u5e38\u7d27\u51d1\u7684 Embedding \u5411\u91cf\uff0c\u6548\u679c\u663e\u8457\u597d\u4e8e\u4f20\u7edf <a href=\"https:\/\/www.kdjingpai.com\/en\/bm25\/\">BM25<\/a> \u68c0\u7d22\u65b9\u6cd5\uff08\u5728 MIRACL \u57fa\u51c6\u6d4b\u8bd5\u7684\u6240\u6709\u8bed\u8a00\u4e2d\uff09\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\"><code>jina-colbert-v2<\/code>\u00a0\u7684\u8bad\u7ec3\u8bed\u6599\u5e93\u8986\u76d6\u4e86 89 \u79cd\u8bed\u8a00\uff0c\u5305\u62ec 4.5 \u4ebf\u5bf9\u5f31\u76d1\u7763\u7684\u8bed\u4e49\u76f8\u5173\u53e5\u5b50\u5bf9\u3001\u95ee\u7b54\u5bf9\u548c\u67e5\u8be2-\u6587\u6863\u5bf9\u3002\u5176\u4e2d\uff0c\u5176\u4e2d\u4e00\u534a\u7684\u6570\u636e\u662f\u82f1\u8bed\uff0c\u5269\u4e0b\u7684\u90e8\u5206\u5305\u62ec 29 \u79cd\u4e0d\u540c\u7684\u975e\u82f1\u8bed\u8bed\u8a00\uff0c\u6b64\u5916\uff0c\u8fd8\u4f7f\u7528\u4e86 3.0% \u7684\u7f16\u7a0b\u8bed\u8a00\u6570\u636e\u548c 4.3% \u7684\u8de8\u8bed\u8a00\u6570\u636e\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u6211\u4eec\u8fd8\u4e13\u95e8\u5bf9\u00a0<strong>\u963f\u62c9\u4f2f\u8bed\u3001\u4e2d\u6587\u3001\u6cd5\u8bed\u3001\u5fb7\u8bed\u3001\u65e5\u8bed\u3001\u4fc4\u8bed\u3001\u897f\u73ed\u7259\u8bed\u7b49\u4e3b\u6d41\u8bed\u8a00<\/strong>\u00a0\u548c\u00a0<strong>\u7f16\u7a0b\u8bed\u8a00<\/strong>\u00a0\u8fdb\u884c\u4e86\u989d\u5916\u8bad\u7ec3\uff0c\u901a\u8fc7\u5f15\u5165\u5bf9\u9f50\u7684\u53cc\u8bed\u6587\u672c\u8bed\u6599\u5e93\uff0c\u4f7f\u5f97\u6a21\u578b\u5728\u5904\u7406\u8de8\u8bed\u8a00\u4efb\u52a1\u65f6\u8868\u73b0\u51fa\u8272\u3002<\/p>\n<figure data-tool=\"mdnice\u7f16\u8f91\u5668\"><figcaption><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter wp-image-5776\" title=\"-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/758656ca1157f2d.png\" alt=\"-1\" width=\"772\" height=\"391\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/758656ca1157f2d.png 1080w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/758656ca1157f2d-300x152.png 300w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/758656ca1157f2d-1024x519.png 1024w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/758656ca1157f2d-768x389.png 768w\" sizes=\"auto, (max-width: 772px) 100vw, 772px\" \/>\u9884\u8bad\u7ec3\u6570\u636e\u96c6\u7684\u8bed\u8a00\u5206\u5e03<\/figcaption><\/figure>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u4e0b\u56fe\u5c55\u793a\u4e86\u5728 MIRACL \u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0cJina-ColBERT-v2 \u548c BM25 \u5728 16 \u79cd\u8bed\u8a00\u4e0a\u7684\u6027\u80fd\u5bf9\u6bd4\u7ed3\u679c\u3002<\/p>\n<h3 data-tool=\"mdnice\u7f16\u8f91\u5668\"><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter wp-image-5777\" title=\"-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/f193d4a175c09c6.png\" alt=\"-1\" width=\"783\" height=\"479\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/f193d4a175c09c6.png 1080w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/f193d4a175c09c6-300x184.png 300w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/f193d4a175c09c6-1024x627.png 1024w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/f193d4a175c09c6-768x470.png 768w\" sizes=\"auto, (max-width: 783px) 100vw, 783px\" \/><\/h3>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\n<h3 data-tool=\"mdnice\u7f16\u8f91\u5668\">3. \u4fc4\u7f57\u65af\u5957\u5a03\u8868\u5f81\u5b66\u4e60<\/h3>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\"><strong>\u4fc4\u7f57\u65af\u5957\u5a03\u8868\u5f81\u5b66\u4e60(Matryoshka Representation Learning, MRL)<\/strong>\u00a0\u662f\u4e00\u79cd\u7075\u6d3b\u4e14\u9ad8\u6548\u7684\u8bad\u7ec3\u65b9\u6cd5\uff0c\u80fd\u591f\u5728\u652f\u6301\u4e0d\u540c\u8f93\u51fa\u5411\u91cf\u7ef4\u5ea6\u7684\u540c\u65f6\uff0c\u6700\u5927\u9650\u5ea6\u5730\u51cf\u5c11\u7cbe\u5ea6\u635f\u5931\u3002\u5177\u4f53\u5b9e\u73b0\u662f\u5728\u6a21\u578b\u7684\u9690\u85cf\u5c42\u4e2d\u5f15\u5165\u591a\u4e2a\u7ebf\u6027\u5c42\uff0c\u6bcf\u4e00\u5c42\u5bf9\u4e0d\u540c\u7ef4\u5ea6\u8fdb\u884c\u5904\u7406\u3002\u8be5\u6280\u672f\u66f4\u591a\u7ec6\u8282\uff0c\u8bf7\u53c2\u9605\uff1ahttps:\/\/arxiv.org\/abs\/2205.13147<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u9ed8\u8ba4\u60c5\u51b5\u4e0b\uff0c<strong>Jina-ColBERT-v2 \u4f1a\u751f\u6210 128 \u7ef4\u7684 Embedding\uff0c\u4f46\u4f60\u53ef\u4ee5\u9009\u62e9\u66f4\u77ed\u7684\u5411\u91cf\uff0c\u6bd4\u5982 96 \u7ef4\u6216 64 \u7ef4<\/strong>\u3002\u5373\u4fbf\u5411\u91cf\u7f29\u77ed\u4e86 25% \u6216 50%\uff0c\u5b83\u5bf9\u6027\u80fd\u7684\u5f71\u54cd\u51e0\u4e4e\u53ef\u4ee5\u5ffd\u7565\uff08\u4e0b\u964d\u5c0f\u4e8e 1.5%\uff09\u3002\u4e5f\u5c31\u662f\u8bf4\uff0c\u65e0\u8bba\u4f60\u9700\u8981\u591a\u957f\u7684\u5411\u91cf\uff0c\u8fd9\u4e2a\u6a21\u578b\u90fd\u80fd\u4fdd\u6301\u9ad8\u6548\u800c\u7cbe\u51c6\u7684\u8868\u73b0\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u4e0b\u8868\u5c55\u793a\u4e86 Jina-ColBERT-v2 \u5728\u8fd9\u516d\u4e2a BEIR \u57fa\u51c6\u6d4b\u8bd5\u4e2d\u524d\u5341\u540d\u7ed3\u679c (nDGC@10) \u7684\u8868\u73b0\u3002\u53ef\u4ee5\u770b\u5230\uff0c<strong>128 \u7ef4\u548c 96 \u7ef4\u4e4b\u95f4\u7684\u6027\u80fd\u5dee\u5f02\u4e0d\u5230 1%\uff0c\u800c 128 \u7ef4\u548c 64 \u7ef4\u4e4b\u95f4\u7684\u5dee\u5f02\u4e0d\u5230 1.5%\u3002<\/strong><\/p>\n<section data-tool=\"mdnice\u7f16\u8f91\u5668\">\n<table>\n<thead>\n<tr>\n<th>\u8f93\u51fa\u7ef4\u5ea6<\/th>\n<th>\u5e73\u5747\u5f97\u5206\uff086 \u9879\u57fa\u51c6\u6d4b\u8bd5\u7684 nDGC@10\uff09<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>128<\/td>\n<td>0.565<\/td>\n<\/tr>\n<tr>\n<td>96<\/td>\n<td>0.558<\/td>\n<\/tr>\n<tr>\n<td>64<\/td>\n<td>0.556<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/section>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u4e0b\u56fe\u76f4\u89c2\u5730\u5c55\u793a\u4e86 Jina-ColBERT-v2 \u5728\u4e0d\u540c\u8f93\u51fa\u7ef4\u5ea6\u4e0b\u7684\u6027\u80fd\u8868\u73b0\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\"><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter wp-image-5778\" title=\"-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/9830bb7f0988b8e.png\" alt=\"-1\" width=\"746\" height=\"543\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/9830bb7f0988b8e.png 1080w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/9830bb7f0988b8e-300x218.png 300w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/9830bb7f0988b8e-1024x745.png 1024w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/09\/9830bb7f0988b8e-768x559.png 768w\" sizes=\"auto, (max-width: 746px) 100vw, 746px\" \/>\u9009\u62e9\u8f83\u5c0f\u7684\u8f93\u51fa\u5411\u91cf\u4e0d\u4ec5\u80fd\u8282\u7701\u5b58\u50a8\u7a7a\u95f4\uff0c\u8fd8\u80fd\u63d0\u5347\u8ba1\u7b97\u901f\u5ea6\uff0c\u5c24\u5176\u662f\u5728\u9700\u8981\u6bd4\u8f83\u5411\u91cf\u6216\u8ba1\u7b97\u8ddd\u79bb\u7684\u573a\u666f\u4e2d\uff0c\u5982\u5411\u91cf\u68c0\u7d22\u7cfb\u7edf\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u540c\u65f6\u6210\u672c\u4e5f\u662f\u975e\u5e38\u660e\u663e\u4e0b\u964d\u7684\uff0c\u6839\u636e Qdrant \u7684\u4e91\u7aef\u6210\u672c\u4f30\u7b97\u5668\uff0c\u5728 AWS \u4e0a\u5b58\u50a8 1 \u4ebf\u4e2a\u6587\u6863\u65f6\uff0c\u4f7f\u7528 128 \u7ef4\u5411\u91cf\u7684\u8d39\u7528\u662f\u6bcf\u6708 1319.24 \u7f8e\u5143\uff0c\u800c\u5982\u679c\u4f7f\u7528 64 \u7ef4\u5411\u91cf\uff0c\u8d39\u7528\u53ef\u4ee5\u76f4\u63a5\u51cf\u534a\uff0c\u964d\u5230\u6bcf\u6708 659.62 \u7f8e\u5143\u3002\u8fd9\u6837\u4e00\u6765\uff0c\u4e0d\u4ec5\u901f\u5ea6\u63d0\u5347\u4e86\uff0c\u6210\u672c\u4e5f\u5927\u5e45\u964d\u4f4e\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\n<h2 data-tool=\"mdnice\u7f16\u8f91\u5668\">\u5982\u4f55\u5f00\u59cb\u4f7f\u7528 Jina ColBERT v2<\/h2>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">Jina ColBERT v2 \u5df2\u7ecf\u4e0a\u7ebf Jina Search Foundation API\u3001AWS Marketplace \u548c Azure \u5e73\u53f0\u3002\u6b64\u5916\uff0c\u5b83\u8fd8\u4ee5 CC BY-NC-4.0 \u8bb8\u53ef\u8bc1\u5728\u00a0<span class=\"wx_search_keyword_wrap\">Hugging Face<i class=\"wx_search_keyword\"><\/i><\/span>\u00a0\u4e0a\u5f00\u6e90\uff0c\u4f9b\u975e\u5546\u4e1a\u7528\u9014\u4f7f\u7528\u3002<\/p>\n<ul class=\"list-paddingleft-1\" data-tool=\"mdnice\u7f16\u8f91\u5668\">\n<li>\n<section>AWS Marketplace: https:\/\/aws.amazon.com\/marketplace\/seller-profile?id=seller-stch2ludm6vgy<\/section>\n<\/li>\n<li>\n<section>Azure: https:\/\/azuremarketplace.microsoft.com\/en-gb\/marketplace\/apps?search=Jina<\/section>\n<\/li>\n<li>\n<section>Hugging Face: https:\/\/huggingface.co\/jinaai\/jina-colbert-v2<\/section>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h3 data-tool=\"mdnice\u7f16\u8f91\u5668\">\u901a\u8fc7 Jina Search Foundation API<\/h3>\n<h4 data-tool=\"mdnice\u7f16\u8f91\u5668\">\u7528\u4e8e Embedding<\/h4>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u4f7f\u7528 Jina Embedding API \u83b7\u53d6 jina-colbert-v2 \u7684 Embedding \u662f\u6700\u7b80\u5355\u76f4\u63a5\u7684\u65b9\u5f0f\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\"><strong>\u4e24\u4e2a\u5173\u952e\u53c2\u6570\uff1a<\/strong><\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\"><code>dimensions<\/code>\uff1a\u8868\u793a Embedding \u7684\u8f93\u51fa\u7ef4\u5ea6\uff0c\u9ed8\u8ba4\u4e3a 128\uff0c\u4e5f\u53ef\u4ee5\u9009\u62e9 64\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\"><code>input_type<\/code>\uff1a\u51b3\u5b9a\u8f93\u5165\u7c7b\u578b\uff0c<code>query<\/code>\u00a0\u7684\u6700\u5927\u957f\u5ea6\u9650\u5236\u4e3a 32 token\uff0c\u8d85\u8fc7\u90e8\u5206\u4f1a\u81ea\u52a8\u622a\u65ad\uff0c\u800c\u00a0<code>document<\/code>\u00a0\u53ef\u4ee5\u957f\u8fbe 8192 token\u3002\u6587\u6863\u7f16\u7801\u53ef\u4ee5\u79bb\u7ebf\u5904\u7406\uff0c\u8fd9\u6837\u5728\u67e5\u8be2\u65f6\u53ea\u9700\u8981\u7f16\u7801\u67e5\u8be2\u90e8\u5206\uff0c\u80fd\u591f\u663e\u8457\u63d0\u9ad8\u5904\u7406\u901f\u5ea6\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">Jina API Key \u5728 jina.ai\/embeddings \u83b7\u53d6\u3002<\/p>\n<pre data-tool=\"mdnice\u7f16\u8f91\u5668\"><code>curl\u00a0https:\/\/api.jina.ai\/v1\/multi-vector\u00a0\\\\\r\n\u00a0\u00a0\u00a0\u00a0\u00a0-H\u00a0\"Content-Type:\u00a0application\/json\"\u00a0\\\\\r\n\u00a0\u00a0\u00a0\u00a0\u00a0-H\u00a0\"Authorization:\u00a0Bearer\u00a0&lt;YOUR\u00a0JINA\u00a0API\u00a0KEY&gt;\"\u00a0\\\\\r\n\u00a0\u00a0\u00a0\u00a0\u00a0-d\u00a0'{\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\"model\":\u00a0\"jina-colbert-v2\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\"dimensions\":\u00a0128,\u00a0#\u00a0\u6216\u00a064\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\"input_type\":\u00a0\"document\",\u00a0#\u00a0\u8fd9\u91cc\u9700\u8981\u6ce8\u610finput_type\u00a0\u662f\u00a0document\u00a0\u8fd8\u662f\u00a0query\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\"embedding_type\":\u00a0\"float\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\"input\":\u00a0[\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"\u5728\u6b64\u8f93\u5165\u4f60\u7684\u6587\u6863\u6587\u672c\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"\u4f60\u53ef\u4ee5\u53d1\u9001\u591a\u4e2a\u6587\u672c\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"\u6bcf\u4e2a\u6587\u672c\u6700\u957f\u53ef\u8fbe\u00a08192\u00a0\u4e2a\u6807\u8bb0\"\r\n\u00a0\u00a0\u00a0\u00a0]}'\r\n<\/code><\/pre>\n<h4 data-tool=\"mdnice\u7f16\u8f91\u5668\">\u7528\u4e8e Reranker<\/h4>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u8981\u901a\u8fc7 Jina Reranker API \u4f7f\u7528\u00a0<code>jina-colbert-v2<\/code>\uff0c\u4f20\u5165\u4e00\u4e2a\u67e5\u8be2\u548c\u591a\u4e2a\u6587\u6863\uff0c\u5e76\u8fd4\u56de\u53ef Reranker \u7684\u5339\u914d\u5206\u6570\uff0c\u6784\u9020\u5982\u4e0b\u8bf7\u6c42\uff1a<\/p>\n<pre data-tool=\"mdnice\u7f16\u8f91\u5668\"><code>curl\u00a0https:\/\/api.jina.ai\/v1\/rerank\u00a0\\\\\r\n\u00a0\u00a0\u00a0\u00a0\u00a0-H\u00a0\"Content-Type:\u00a0application\/json\"\u00a0\\\\\r\n\u00a0\u00a0\u00a0\u00a0\u00a0-H\u00a0\"Authorization:\u00a0Bearer\u00a0&lt;YOUR\u00a0JINA\u00a0API\u00a0KEY&gt;\"\u00a0\\\\\r\n\u00a0\u00a0\u00a0\u00a0\u00a0-d\u00a0'{\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"model\":\u00a0\"jina-colbert-v2\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"query\":\u00a0\"\u67cf\u6797\u7684\u4eba\u53e3\u662f\u591a\u5c11\uff1f\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"top_n\":\u00a03,\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"documents\":\u00a0[\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"2023\u5e74\u67cf\u6797\u7684\u4eba\u53e3\u6bd4\u53bb\u5e74\u589e\u957f\u4e860.7%\u3002\u56e0\u6b64\uff0c\u5230\u53bb\u5e74\u5e74\u5e95\uff0c\u67cf\u6797\u7684\u5c45\u6c11\u4eba\u6570\u6bd42022\u5e74\u589e\u52a0\u4e86\u7ea627300\u4eba\u300230\u5c81\u523040\u5c81\u7684\u4eba\u7fa4\u662f\u6570\u91cf\u6700\u591a\u7684\u5e74\u9f84\u7ec4\u3002\u67cf\u6797\u62e5\u6709\u7ea688.1\u4e07\u5916\u56fd\u5c45\u6c11\uff0c\u6765\u81ea\u7ea6170\u4e2a\u56fd\u5bb6\uff0c\u5e73\u5747\u5e74\u9f84\u4e3a42.5\u5c81\u3002\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"\u67cf\u6797\u5c71\u662f\u5357\u6781\u6d32\u739b\u4e3d\u00b7\u4f2f\u5fb7\u5730\u7684\u4e00\u4e2a\u51b0\u5ddd\u8986\u76d6\u7684\u706b\u5c71\uff0c\u8ddd\u79bb\u963f\u8499\u68ee\u6d77\u7ea6100\u516c\u91cc\uff0862\u82f1\u91cc\uff09\u3002\u5b83\u662f\u4e00\u4e2a\u5bbd\u7ea620\u516c\u91cc\uff0812\u82f1\u91cc\uff09\u7684\u5c71\u5cf0\uff0c\u6709\u5bc4\u751f\u706b\u5c71\u53e3\uff0c\u7531\u4e24\u5ea7\u5408\u5e76\u7684\u706b\u5c71\u7ec4\u6210\uff1a\u67cf\u6797\u706b\u5c71\uff0c\u5176\u706b\u5c71\u53e3\u5bbd2\u516c\u91cc\uff081.2\u82f1\u91cc\uff09\uff0c\u4ee5\u53ca\u8ddd\u79bb\u67cf\u6797\u706b\u5c71\u7ea63.5\u516c\u91cc\uff082.2\u82f1\u91cc\uff09\u7684Merrem\u5cf0\uff0c\u706b\u5c71\u53e3\u5bbd2.5\u00d71\u516c\u91cc\uff081.55\u82f1\u91cc\u00d70.62\u82f1\u91cc\uff09\u3002\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"\u622a\u81f32023\u5e7412\u670831\u65e5\uff0c\u5404\u8054\u90a6\u5dde\u6309\u56fd\u7c4d\u548c\u5dde\u7edf\u8ba1\u7684\u4eba\u53e3\u6570\u636e\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"\u67cf\u6797\u7684\u90fd\u5e02\u533a\u4eba\u53e3\u8d85\u8fc7450\u4e07\uff0c\u662f\u5fb7\u56fd\u4eba\u53e3\u6700\u591a\u7684\u90fd\u5e02\u533a\u3002\u67cf\u6797-\u52c3\u5170\u767b\u5821\u9996\u90fd\u5730\u533a\u7ea6\u6709620\u4e07\u4eba\u53e3\uff0c\u662f\u5fb7\u56fd\u7b2c\u4e8c\u5927\u90fd\u5e02\u533a\uff0c\u4ec5\u6b21\u4e8e\u83b1\u8335-\u9c81\u5c14\u533a\uff0c\u5e76\u4e14\u662f\u6b27\u76df\u7b2c\u516d\u5927\u90fd\u5e02\u533a\uff08\u6309GDP\u8ba1\u7b97\uff09\u3002\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"\u6b27\u6587\u00b7\u67cf\u6797\uff08\u539f\u540d\u4ee5\u8272\u5217\u00b7\u8d1d\u6797\uff09\u662f\u4e00\u4f4d\u7f8e\u56fd\u4f5c\u66f2\u5bb6\u548c\u8bcd\u66f2\u4f5c\u8005\u3002\u4ed6\u7684\u97f3\u4e50\u662f\u300a\u4f1f\u5927\u7684\u7f8e\u56fd\u6b4c\u672c\u300b\u7684\u4e00\u90e8\u5206\u3002\u67cf\u6797\u83b7\u5f97\u4e86\u8bb8\u591a\u8363\u8a89\uff0c\u5305\u62ec\u5965\u65af\u5361\u5956\u3001\u683c\u83b1\u7f8e\u5956\u548c\u6258\u5c3c\u5956\u3002\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"\u67cf\u6797\u662f\u7f8e\u56fd\u5eb7\u6d85\u72c4\u683c\u5dde\u9996\u90fd\u89c4\u5212\u533a\u7684\u4e00\u4e2a\u57ce\u9547\u30022020\u5e74\u4eba\u53e3\u666e\u67e5\u65f6\u4eba\u53e3\u4e3a20,175\u3002\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"\u67cf\u6797\u662f\u5fb7\u56fd\u7684\u9996\u90fd\u548c\u6700\u5927\u57ce\u5e02\uff0c\u65e0\u8bba\u662f\u9762\u79ef\u8fd8\u662f\u4eba\u53e3\u3002\u5176\u8d85\u8fc7385\u4e07\u5c45\u6c11\u4f7f\u5176\u6210\u4e3a\u6b27\u76df\u4eba\u53e3\u6700\u591a\u7684\u57ce\u5e02\uff08\u6309\u57ce\u5e02\u9650\u754c\u5185\u4eba\u53e3\u8ba1\u7b97\uff09\u3002\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"\u300a\u67cf\u6797\uff0c\u67cf\u6797\u300b\u662f\u4e00\u90e8\u4e3aARD\u5236\u4f5c\u7684\u7535\u89c6\u5267\uff0c\u4e8e2002\u5e74\u81f32005\u5e74\u5728\u5fb7\u56fd\u7b2c\u4e00\u7535\u89c6\u53f0\u7684\u665a\u95f4\u8282\u76ee\u4e2d\u64ad\u51fa\u3002\u5bfc\u6f14\u5305\u62ec\u5f17\u6717\u8328\u65af\u5361\u00b7\u8fc8\u8036\u00b7\u666e\u83b1\u65af\u3001\u514b\u91cc\u65af\u6258\u592b\u00b7\u65bd\u5185\u3001\u65af\u6587\u00b7\u6e29\u7279\u74e6\u5c14\u7279\u548c\u8482\u56fe\u65af\u00b7\u585e\u5c14\u683c\u3002\"\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0]\r\n\u00a0\u00a0\u00a0\u00a0}'\r\n<\/code><\/pre>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u8fd8\u8981\u6ce8\u610f\u00a0<code>top_n<\/code>\u00a0\u53c2\u6570\uff0c\u5b83\u6307\u5b9a\u68c0\u7d22\u65f6\u8fd4\u56de\u7684\u6587\u6863\u6570\u91cf\u3002\u5982\u679c\u4f60\u53ea\u9700\u8981\u6700\u4f18\u7684\u5339\u914d\u7ed3\u679c\uff0c\u53ef\u4ee5\u628a\u00a0<code>top_n<\/code>\u00a0\u8bbe\u7f6e\u4e3a 1\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u6709\u5173\u66f4\u591a\u7684 Python \u6216\u5176\u4ed6\u8bed\u8a00\u7684\u793a\u4f8b\u4ee3\u7801\uff0c\u8bf7\u8bbf\u95ee https:\/\/jina.ai\/embeddings \u9875\u9762\uff0c\u6216\u5728 https:\/\/jina.ai\/reranker\/ \u7684\u4e0b\u62c9\u83dc\u5355\u4e2d\u9009\u62e9\u00a0<code>jina-colbert-v2<\/code>\u3002<\/p>\n<h3 data-tool=\"mdnice\u7f16\u8f91\u5668\">\u901a\u8fc7 Stanford ColBERT<\/h3>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u5982\u679c\u4f60\u7528\u8fc7 Stanford ColBERT \u5e93\uff0c\u4f60\u73b0\u5728\u53ef\u4ee5\u65e0\u7f1d\u66ff\u6362\u6210 Jina ColBERT v2 \u7248\u672c\u3002\u53ea\u9700\u6307\u5b9a\u00a0<code>jinaai\/jina-colbert-v2<\/code>\u00a0\u4f5c\u4e3a\u6a21\u578b\u6e90\u3002<\/p>\n<pre data-tool=\"mdnice\u7f16\u8f91\u5668\"><code>from\u00a0colbert.infra\u00a0import\u00a0ColBERTConfig\r\nfrom\u00a0colbert.modeling.checkpoint\u00a0import\u00a0Checkpoint\r\n\r\nckpt\u00a0=\u00a0Checkpoint(\"jinaai\/jina-colbert-v2\",\u00a0colbert_config=ColBERTConfig())\r\ndocs\u00a0=\u00a0[\"\u4f60\u7684\u6587\u672c\u5217\u8868\"]\r\nquery_vectors\u00a0=\u00a0ckpt.queryFromText(docs)\r\n<\/code><\/pre>\n<h3 data-tool=\"mdnice\u7f16\u8f91\u5668\">\u901a\u8fc7 RAGatouille<\/h3>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">Jina ColBERT v2 \u540c\u6837\u96c6\u6210\u5728 RAGatouille \u7cfb\u7edf\u4e2d\uff0c\u901a\u8fc7\u00a0<code>RAGPretrainedModel.from_pretrained()<\/code>\u00a0\u65b9\u6cd5\u5c31\u80fd\u4e0b\u8f7d\u4f7f\u7528\uff0c\u975e\u5e38\u65b9\u4fbf\u3002<\/p>\n<pre data-tool=\"mdnice\u7f16\u8f91\u5668\"><code>from\u00a0ragatouille\u00a0import\u00a0RAGPretrainedModel\r\n\r\nRAG\u00a0=\u00a0RAGPretrainedModel.from_pretrained(\"jinaai\/jina-colbert-v2\")\r\ndocs\u00a0=\u00a0[\"\u4f60\u7684\u6587\u672c\u5217\u8868\"]\r\nRAG.index(docs,\u00a0index_name=\"your_index_name\")\r\nquery\u00a0=\u00a0\"\u4f60\u7684\u67e5\u8be2\"\r\nresults\u00a0=\u00a0RAG.search(query)\r\n<\/code><\/pre>\n<h3 data-tool=\"mdnice\u7f16\u8f91\u5668\">\u901a\u8fc7 Qdrant<\/h3>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u4ece Qdrant 1.10 \u7248\u672c\u5f00\u59cb\uff0cQdrant \u652f\u6301\u4e86\u591a\u5411\u91cf\u548c\u8fdf\u4ea4\u4e92\u6a21\u578b\uff0c\u4f60\u53ef\u4ee5\u76f4\u63a5\u4f7f\u7528 jina-colbert-v2\u3002\u4e0d\u8bba\u662f\u672c\u5730\u90e8\u7f72\u8fd8\u662f\u4e91\u7aef\u6258\u7ba1\u7684 Qdrant\uff0c\u53ea\u9700\u8981\u5728\u5ba2\u6237\u7aef\u4e2d\u6b63\u786e\u914d\u7f6e multivector_config \u53c2\u6570\uff0c\u5c31\u53ef\u4ee5\u628a\u6587\u6863\u63d2\u5165\u5230\u591a\u5411\u91cf\u96c6\u5408\u4e2d\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\"><strong>\u4f7f\u7528 MAX_SIM \u64cd\u4f5c\u521b\u5efa\u65b0\u96c6\u5408<\/strong><\/p>\n<pre data-tool=\"mdnice\u7f16\u8f91\u5668\"><code>from\u00a0qdrant_client\u00a0import\u00a0QdrantClient,\u00a0models\r\n\r\nqdrant_client\u00a0=\u00a0QdrantClient(\r\n\u00a0\u00a0\u00a0\u00a0url=\"&lt;YOUR_ENDPOINT&gt;\",\r\n\u00a0\u00a0\u00a0\u00a0api_key=\"&lt;YOUR_API_KEY&gt;\",\r\n)\r\n\r\nqdrant_client.create_collection(\r\n\u00a0\u00a0\u00a0\u00a0collection_name=\"{collection_name}\",\r\n\u00a0\u00a0\u00a0\u00a0vectors_config={\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"colbert\":\u00a0models.VectorParams(\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0size=128,\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0distance=models.Distance.COSINE,\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0multivector_config=models.MultiVectorConfig(\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0comparator=models.MultiVectorComparator.MAX_SIM\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0),\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0)\r\n\u00a0\u00a0\u00a0\u00a0}\r\n)\r\n<\/code><\/pre>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">\u26a0\ufe0f \u6b63\u786e\u8bbe\u7f6e\u00a0<code>multivector_config<\/code>\u00a0\u53c2\u6570\u662f Qdrant \u4e2d\u4f7f\u7528 ColBERT \u6a21\u578b\u7684\u5173\u952e\u3002<\/p>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\"><strong>\u5c06\u6587\u6863\u63d2\u5165\u591a\u5411\u91cf\u96c6\u5408<\/strong><\/p>\n<pre data-tool=\"mdnice\u7f16\u8f91\u5668\"><code>import\u00a0requests\r\nfrom\u00a0qdrant_client\u00a0import\u00a0QdrantClient,\u00a0models\r\n\r\nurl\u00a0=\u00a0'https:\/\/api.jina.ai\/v1\/multi-vector'\r\n\r\nheaders\u00a0=\u00a0{\r\n\u00a0\u00a0\u00a0\u00a0'Content-Type':\u00a0'application\/json',\r\n\u00a0\u00a0\u00a0\u00a0'Authorization':\u00a0'Bearer\u00a0&lt;YOUR\u00a0BEARER&gt;'\r\n}\r\n\r\ndata\u00a0=\u00a0{\r\n\u00a0\u00a0\u00a0\u00a0'model':\u00a0'jina-colbert-v2',\r\n\u00a0\u00a0\u00a0\u00a0'input_type':\u00a0'query',\r\n\u00a0\u00a0\u00a0\u00a0'embedding_type':\u00a0'float',\r\n\u00a0\u00a0\u00a0\u00a0'input':\u00a0[\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0'\u5728\u6b64\u8f93\u5165\u4f60\u7684\u6587\u672c',\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0'\u4f60\u53ef\u4ee5\u53d1\u9001\u591a\u4e2a\u6587\u672c',\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0'\u6bcf\u4e2a\u6587\u672c\u6700\u957f\u53ef\u8fbe\u00a08192\u00a0\u4e2a\u6807\u8bb0'\r\n\u00a0\u00a0\u00a0\u00a0]\r\n}\r\n\r\nresponse\u00a0=\u00a0requests.post(url,\u00a0headers=headers,\u00a0json=data)\r\n<a href=\"https:\/\/www.kdjingpai.com\/en\/rowsshujuqudongai\/\">rows<\/a>\u00a0=\u00a0response.json()[\"data\"]\r\n\r\nqdrant_client\u00a0=\u00a0QdrantClient(\r\n\u00a0\u00a0\u00a0\u00a0url=\"&lt;YOUR_ENDPOINT&gt;\",\r\n\u00a0\u00a0\u00a0\u00a0api_key=\"&lt;YOUR\u00a0API_KEY&gt;\",\r\n)\r\n\r\nfor\u00a0i,\u00a0row\u00a0in\u00a0enumerate(rows):\r\n\u00a0\u00a0\u00a0\u00a0qdrant_client.upsert(\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0collection_name=\"{collection_name}\",\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0points=[\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0models.PointStruct(\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0id=i,\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0vector=row[\"embeddings\"],\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0payload={\"text\":\u00a0data[\"input\"][i]}\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0)\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0],\r\n\u00a0\u00a0\u00a0\u00a0)\r\n<\/code><\/pre>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\"><strong>\u67e5\u8be2\u96c6\u5408<\/strong><\/p>\n<pre data-tool=\"mdnice\u7f16\u8f91\u5668\"><code>from\u00a0qdrant_client\u00a0import\u00a0QdrantClient,\u00a0models\r\nimport\u00a0requests\r\n\r\nurl\u00a0=\u00a0'https:\/\/api.jina.ai\/v1\/multi-vector'\r\n\r\nheaders\u00a0=\u00a0{\r\n\u00a0\u00a0\u00a0\u00a0'Content-Type':\u00a0'application\/json',\r\n\u00a0\u00a0\u00a0\u00a0'Authorization':\u00a0'Bearer\u00a0&lt;YOUR\u00a0BEARER&gt;'\r\n}\r\n\r\ndata\u00a0=\u00a0{\r\n\u00a0\u00a0\u00a0\u00a0'model':\u00a0'jina-colbert-v2',\r\n\u00a0\u00a0\u00a0\u00a0\"input_type\":\u00a0\"query\",\r\n\u00a0\u00a0\u00a0\u00a0\"embedding_type\":\u00a0\"float\",\r\n\u00a0\u00a0\u00a0\u00a0\"input\":\u00a0[\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"Jina AI \u7684Embedding\u6a21\u578b\u652f\u6301\u591a\u5c11\u4e2a\u6807\u8bb0\u7684\u8f93\u5165\uff1f\"\r\n\u00a0\u00a0\u00a0\u00a0]\r\n}\r\n\r\nresponse\u00a0=\u00a0requests.post(url,\u00a0headers=headers,\u00a0json=data)\r\nvector\u00a0=\u00a0response.json()[\"data\"][0][\"embeddings\"]\r\n\r\nqdrant_client\u00a0=\u00a0QdrantClient(\r\n\u00a0\u00a0\u00a0\u00a0url=\"&lt;YOUR_ENDPOINT&gt;\",\r\n\u00a0\u00a0\u00a0\u00a0api_key=\"&lt;YOUR\u00a0API_KEY&gt;\",\r\n)\r\n\r\nresults\u00a0=\u00a0qdrant_client.query_points(\r\n\u00a0\u00a0\u00a0\u00a0collection_name=\"{collection_name}\",\r\n\u00a0\u00a0\u00a0\u00a0query=vector,\r\n)\r\n\r\nprint(results)\r\n<\/code><\/pre>\n<h3 data-tool=\"mdnice\u7f16\u8f91\u5668\">\u603b\u7ed3<\/h3>\n<p data-tool=\"mdnice\u7f16\u8f91\u5668\">jina-colbert-v2 \u5728\u524d\u4ee3\u7248\u672c\u7684\u57fa\u7840\u4e0a\uff0c\u8986\u76d6\u4e86 89 \u79cd\u5168\u7403\u8bed\u8a00\uff0c\u4ee5\u53ca\u591a\u79cd Embedding \u8f93\u51fa\u7ef4\u5ea6\u7684\u9009\u62e9\uff0c\u7528\u6237\u80fd\u591f\u7075\u6d3b\u5e73\u8861\u7cbe\u5ea6\u548c\u6548\u7387\u9700\u6c42\uff0c\u8282\u7701\u8ba1\u7b97\u548c\u65f6\u95f4\u6210\u672c\u3002\u70b9\u51fb https:\/\/jina.ai \u7acb\u5373\u8bd5\u7528\uff0c\u8fd8\u53ef\u83b7\u5f97 100 \u4e07\u4e2a\u514d\u8d39 Token\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5728\u00a0RAG\u00a0\u9886\u57df\uff0c\u591a\u5411\u91cf\u6a21\u578b\u00a0ColBERT\u00a0\u901a\u8fc7\u4e3a\u6587\u6863\u7684\u6bcf\u4e2a token \u751f\u6210\u72ec\u7acb\u7684\u5411\u91cf\uff0c\u5e26\u6765\u4e86\u68c0\u7d22\u7cbe\u5ea6\u7684\u63d0\u5347\u3002\u4f46\u540c\u6837\u4e5f\u5e26\u6765\u4e86\u5b58\u50a8\u9700\u6c42\u7684\u5267\u589e\uff0c\u5e76\u4e14\u4ec5\u652f\u6301\u82f1\u6587\uff0c\u9650\u5236\u4e86\u5176\u5e94\u7528\u8303\u56f4\u3002 \u4e3a\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\uff0c\u6211\u4eec\u6539\u8fdb\u4e86 ColBERT \u7684\u67b6\u6784\u548c\u8bad\u7ec3\u6d41\u7a0b&#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[46],"tags":[],"class_list":["post-5773","post","type-post","status-publish","format-standard","hentry","category-news"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts\/5773","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/comments?post=5773"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts\/5773\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/media?parent=5773"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/categories?post=5773"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/tags?post=5773"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}