{"id":92884,"date":"2026-08-11T13:56:48","date_gmt":"2026-08-11T05:56:48","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/92884.html"},"modified":"2026-08-11T13:56:48","modified_gmt":"2026-08-11T05:56:48","slug":"%e5%a4%9a%e6%a8%a1%e6%80%81%e8%ae%ad%e7%bb%83%ef%bc%9a%e4%bb%8e%e5%83%8f%e7%b4%a0%e5%88%b0%e8%af%ad%e4%b9%89%ef%bc%8c%e4%b8%80%e4%b8%aa%e6%a8%a1%e5%9e%8b%e7%bb%9f%e6%b2%bb%e6%89%80%e6%9c%89%e6%a8%a1","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/92884.html","title":{"rendered":"\u591a\u6a21\u6001\u8bad\u7ec3\uff1a\u4ece\u50cf\u7d20\u5230\u8bed\u4e49\uff0c\u4e00\u4e2a\u6a21\u578b\u7edf\u6cbb\u6240\u6709\u6a21\u6001"},"content":{"rendered":"<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260811055646-6a7ab99ee25b3.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"><\/p>\n<h3>\u76ee\u5f55<\/h3>\n<li>\u591a\u6a21\u6001\u8bad\u7ec3\u6982\u8ff0<\/li>\n<li>\u89c6\u89c9\u7f16\u7801\u5668\uff1aViT \u4e0e CLIP<\/li>\n<li>\u591a\u6a21\u6001\u5bf9\u9f50\uff1a\u5bf9\u6bd4\u5b66\u4e60\u4e0e\u8de8\u6a21\u6001\u6ce8\u610f\u529b<\/li>\n<li>\u591a\u6a21\u6001\u878d\u5408\uff1a\u65e9\u671f\u878d\u5408\u4e0e\u665a\u671f\u878d\u5408<\/li>\n<li>\u591a\u6a21\u6001\u751f\u6210\uff1a\u6587\u672c\u5230\u56fe\u50cf\u3001\u56fe\u50cf\u5230\u6587\u672c<\/li>\n<li>\u591a\u6a21\u6001\u8bad\u7ec3\u6311\u6218<\/li>\n<li>\u5b9e\u8df5\u5efa\u8bae\u4e0e\u6700\u4f73\u5b9e\u8df5<\/li>\n<h3>\u6458\u8981<\/h3>\n<p>\u591a\u6a21\u6001\u8bad\u7ec3\u628a\u6587\u672c\u3001\u56fe\u50cf\u3001\u97f3\u9891\u6295\u5f71\u5230\u540c\u4e00\u5411\u91cf\u7a7a\u95f4\uff0c\u8ba9\u4e00\u4e2a 7B \u89c4\u6a21 LLM \u540c\u65f6\u627f\u62c5\u7406\u89e3\u4e0e\u751f\u6210\u4e24\u7c7b\u4efb\u52a1\u3002\u4ee5 CLIP-L \u89c6\u89c9\u7f16\u7801\u5668\u8f93\u51fa 1024 \u7ef4\u7279\u5f81\u3001\u5355\u5c42\u7ebf\u6027\u6295\u5f71\u6620\u5c04\u5230 4096 \u7ef4\u8bcd\u5d4c\u5165\u7684\u914d\u7f6e\u4e3a\u4f8b\uff0c\u6295\u5f71\u4ec5\u589e\u52a0\u7ea6 4.2M \u53c2\u6570\uff0c\u8bad\u7ec3\u663e\u5b58\u76f8\u6bd4\u7eaf\u6587\u672c\u591a 30% \u5230 50%\u3002\u672c\u6587\u6309\u89c6\u89c9\u7f16\u7801\u3001\u5bf9\u6bd4\u5bf9\u9f50\u3001\u878d\u5408\u7b56\u7565\u3001\u751f\u6210\u4efb\u52a1\u4e0e\u8bad\u7ec3\u5de5\u7a0b\u4e94\u6761\u7ebf\u5c55\u5f00\uff0c\u7ed9\u51fa\u53ef\u8fd0\u884c\u7684 PyTorch \u4ee3\u7801\u3001\u5bf9\u6bd4\u8868\u4e0e\u5177\u4f53\u6570\u503c\u3002<\/p>\n<h3>1. \u591a\u6a21\u6001\u8bad\u7ec3\u6982\u8ff0<\/h3>\n<p>\u591a\u6a21\u6001\u8bad\u7ec3\u7684\u76ee\u6807\u662f\u8ba9\u5355\u4e00\u6a21\u578b\u63a5\u6536\u5e76\u5904\u7406\u591a\u79cd\u8f93\u5165\u5f62\u5f0f\uff0c\u5e76\u5728\u540c\u4e00\u7ec4\u53c2\u6570\u4e0a \u540c\u65f6\u5b8c\u6210\u7406\u89e3\u4e0e\u751f\u6210\u3002\u672c\u6587\u4ee5\u6587\u672c\u3001\u56fe\u50cf\u3001\u97f3\u9891\u4e09\u7c7b\u6a21\u6001\u4e3a\u8ba8\u8bba\u8fb9\u754c\uff0c\u9aa8\u5e72\u662f 7B \u53c2\u6570\u7684 decoder-only LLM\uff0c\u89c6\u89c9\u4fa7\u9ed8\u8ba4 CLIP-L \u89c6\u89c9\u7f16\u7801\u5668\uff0c\u97f3\u9891\u4fa7\u9ed8\u8ba4 \u5377\u79ef\u58f0\u8c31\u7f16\u7801\u5668\u3002\u5168\u6587\u6d89\u53ca\u7684\u663e\u5b58\u6570\u503c\u4ee5 BF16 \u6df7\u5408\u7cbe\u5ea6\u3001\u6279\u5927\u5c0f 4\u3001\u5e8f\u5217\u957f\u5ea6 2048 \u4e3a\u9ed8\u8ba4\u914d\u7f6e\uff0c\u8d85\u51fa\u8be5\u914d\u7f6e\u7684\u5730\u65b9\u4f1a\u5355\u72ec\u8bf4\u660e\u3002<\/p>\n<p>  #mermaid-svg-MzMl14JI9fpNkJgh{font-family:\\&#8221;trebuchet ms\\&#8221;,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-MzMl14JI9fpNkJgh .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-MzMl14JI9fpNkJgh .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-MzMl14JI9fpNkJgh .error-icon{fill:#552222;}#mermaid-svg-MzMl14JI9fpNkJgh .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-MzMl14JI9fpNkJgh .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-MzMl14JI9fpNkJgh .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-MzMl14JI9fpNkJgh .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-MzMl14JI9fpNkJgh .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-MzMl14JI9fpNkJgh .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-MzMl14JI9fpNkJgh .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-MzMl14JI9fpNkJgh .marker{fill:#333333;stroke:#333333;}#mermaid-svg-MzMl14JI9fpNkJgh .marker.cross{stroke:#333333;}#mermaid-svg-MzMl14JI9fpNkJgh svg{font-family:\\&#8221;trebuchet ms\\&#8221;,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-MzMl14JI9fpNkJgh p{margin:0;}#mermaid-svg-MzMl14JI9fpNkJgh .label{font-family:\\&#8221;trebuchet ms\\&#8221;,verdana,arial,sans-serif;color:#333;}#mermaid-svg-MzMl14JI9fpNkJgh .cluster-label text{fill:#333;}#mermaid-svg-MzMl14JI9fpNkJgh .cluster-label span{color:#333;}#mermaid-svg-MzMl14JI9fpNkJgh .cluster-label span p{background-color:transparent;}#mermaid-svg-MzMl14JI9fpNkJgh .label text,#mermaid-svg-MzMl14JI9fpNkJgh span{fill:#333;color:#333;}#mermaid-svg-MzMl14JI9fpNkJgh .node rect,#mermaid-svg-MzMl14JI9fpNkJgh .node circle,#mermaid-svg-MzMl14JI9fpNkJgh .node ellipse,#mermaid-svg-MzMl14JI9fpNkJgh .node polygon,#mermaid-svg-MzMl14JI9fpNkJgh .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-MzMl14JI9fpNkJgh .rough-node .label text,#mermaid-svg-MzMl14JI9fpNkJgh .node .label text,#mermaid-svg-MzMl14JI9fpNkJgh .image-shape .label,#mermaid-svg-MzMl14JI9fpNkJgh .icon-shape .label{text-anchor:middle;}#mermaid-svg-MzMl14JI9fpNkJgh .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-MzMl14JI9fpNkJgh .rough-node .label,#mermaid-svg-MzMl14JI9fpNkJgh .node .label,#mermaid-svg-MzMl14JI9fpNkJgh .image-shape .label,#mermaid-svg-MzMl14JI9fpNkJgh .icon-shape .label{text-align:center;}#mermaid-svg-MzMl14JI9fpNkJgh .node.clickable{cursor:pointer;}#mermaid-svg-MzMl14JI9fpNkJgh .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-MzMl14JI9fpNkJgh .arrowheadPath{fill:#333333;}#mermaid-svg-MzMl14JI9fpNkJgh .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-MzMl14JI9fpNkJgh .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-MzMl14JI9fpNkJgh .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-MzMl14JI9fpNkJgh .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-MzMl14JI9fpNkJgh .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-MzMl14JI9fpNkJgh .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-MzMl14JI9fpNkJgh .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-MzMl14JI9fpNkJgh .cluster text{fill:#333;}#mermaid-svg-MzMl14JI9fpNkJgh .cluster span{color:#333;}#mermaid-svg-MzMl14JI9fpNkJgh div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\\&#8221;trebuchet ms\\&#8221;,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-MzMl14JI9fpNkJgh .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-MzMl14JI9fpNkJgh rect.text{fill:none;stroke-width:0;}#mermaid-svg-MzMl14JI9fpNkJgh .icon-shape,#mermaid-svg-MzMl14JI9fpNkJgh .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-MzMl14JI9fpNkJgh .icon-shape p,#mermaid-svg-MzMl14JI9fpNkJgh .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-MzMl14JI9fpNkJgh .icon-shape .label rect,#mermaid-svg-MzMl14JI9fpNkJgh .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-MzMl14JI9fpNkJgh .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-MzMl14JI9fpNkJgh .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-MzMl14JI9fpNkJgh :root{&#8211;mermaid-font-family:\\&#8221;trebuchet ms\\&#8221;,verdana,arial,sans-serif;}#mermaid-svg-MzMl14JI9fpNkJgh .default&gt;*{fill:#faf9f5!important;stroke:#ffffff!important;color:#000000!important;stroke-width:0px!important;}#mermaid-svg-MzMl14JI9fpNkJgh .default span{fill:#faf9f5!important;stroke:#ffffff!important;color:#000000!important;stroke-width:0px!important;}#mermaid-svg-MzMl14JI9fpNkJgh .default tspan{fill:#000000!important;} <\/p>\n<p>         <span class=\"nodeLabel\"><\/p>\n<p>\u4efb\u52a1\u5c42<\/p>\n<p><\/span> <\/p>\n<p>         <span class=\"nodeLabel\"><\/p>\n<p>\u878d\u5408\u5c42<\/p>\n<p><\/span> <\/p>\n<p>         <span class=\"nodeLabel\"><\/p>\n<p>\u7f16\u7801\u5c42<\/p>\n<p><\/span> <\/p>\n<p>         <span class=\"nodeLabel\"><\/p>\n<p>\u6570\u636e\u5c42<\/p>\n<p><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/p>\n<p>token \u5e8f\u5217<\/p>\n<p><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/p>\n<p>\u56fe\u50cf\u5f20\u91cf<\/p>\n<p><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/p>\n<p>\u58f0\u8c31\u56fe<\/p>\n<p><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/p>\n<p>\u635f\u5931\u672a\u6536\u655b<\/p>\n<p><\/span> <\/p>\n<p>         <span class=\"edgeLabel\"><\/p>\n<p>\u8d28\u91cf\u4e0d\u8fbe\u6807<\/p>\n<p><\/span> <\/p>\n<p>         <span style=\"color:#000000 !important\" class=\"nodeLabel\"><\/p>\n<p>\u6587\u672c\u8bed\u6599<\/p>\n<p><\/span> <\/p>\n<p>         <span style=\"color:#000000 !important\" class=\"nodeLabel\"><\/p>\n<p>\u56fe\u50cf\u6837\u672c<\/p>\n<p><\/span> <\/p>\n<p>         <span style=\"color:#000000 !important\" class=\"nodeLabel\"><\/p>\n<p>\u97f3\u9891\u6ce2\u5f62<\/p>\n<p><\/span> <\/p>\n<p>         <span style=\"color:#000000 !important\" class=\"nodeLabel\"><\/p>\n<p>\u6587\u672c\u5206\u8bcd\u5668<\/p>\n<p><\/span> <\/p>\n<p>         <span style=\"color:#000000 !important\" class=\"nodeLabel\"><\/p>\n<p>\u89c6\u89c9\u7f16\u7801\u5668<\/p>\n<p><\/span> <\/p>\n<p>         <span style=\"color:#000000 !important\" class=\"nodeLabel\"><\/p>\n<p>\u97f3\u9891\u7f16\u7801\u5668<\/p>\n<p><\/span> <\/p>\n<p>         <span style=\"color:#000000 !important\" class=\"nodeLabel\"><\/p>\n<p>\u6a21\u6001\u6295\u5f71<\/p>\n<p><\/span> <\/p>\n<p>         <span style=\"color:#000000 !important\" class=\"nodeLabel\"><\/p>\n<p>\u8de8\u6a21\u6001\u6ce8\u610f\u529b<\/p>\n<p><\/span> <\/p>\n<p>         <span style=\"color:#000000 !important\" class=\"nodeLabel\"><\/p>\n<p>\u89c6\u89c9\u95ee\u7b54<\/p>\n<p><\/span> <\/p>\n<p>         <span style=\"color:#000000 !important\" class=\"nodeLabel\"><\/p>\n<p>\u56fe\u6587\u751f\u6210<\/p>\n<p><\/span> <\/p>\n<p>         <span style=\"color:#000000 !important\" class=\"nodeLabel\"><\/p>\n<p>\u8bed\u97f3\u7406\u89e3<\/p>\n<p><\/span> <\/p>\n<h4>1.1 \u591a\u6a21\u6001\u8bad\u7ec3\u7684\u5fc5\u8981\u6027<\/h4>\n<p>\u5355\u6a21\u6001\u6a21\u578b\u5728\u4fe1\u606f\u8986\u76d6\u4e0a\u6709\u786c\u8fb9\u754c\u3002\u7eaf\u6587\u672c\u6a21\u578b\u53ea\u80fd\u8bfb\u5230\u56fe\u7247\u7684\u66ff\u4ee3\u6587\u672c\uff0c\u65e0\u6cd5 \u5229\u7528\u50cf\u7d20\u7ea7\u7ec6\u8282\uff1b\u53ea\u5efa\u6a21\u6587\u672c\u7684\u8bed\u97f3\u7cfb\u7edf\u4f1a\u4e22\u5931\u8bed\u6c14\u3001\u505c\u987f\u4e0e\u80cc\u666f\u566a\u58f0\u7b49\u526f\u8bed\u8a00 \u4fe1\u606f\u3002ImageNet \u7684 100 \u4e07\u5f20\u6807\u6ce8\u56fe\u53ea\u80fd\u8986\u76d6 1000 \u4e2a\u7c7b\u522b\uff0c\u800c\u6587\u5b57\u63cf\u8ff0\u53ef\u4ee5\u5145\u5f53 \u5f00\u653e\u7c7b\u522b\u96c6\u5408\u7684\u68c0\u7d22\u8f74\uff0c\u8fd9\u662f\u56fe\u6587\u5bf9\u9f50\u5e26\u6765\u7684\u7ed3\u6784\u6027\u589e\u76ca\uff0c\u4e0d\u662f\u53c2\u6570\u5806\u53e0\u80fd\u66ff\u4ee3\u7684\u3002<\/p>\n<p>\u5de5\u7a0b\u6536\u76ca\u53ef\u4ee5\u91cf\u5316\u3002\u628a\u5546\u54c1\u56fe\u6587\u5bf9\u5582\u7ed9\u5bf9\u6bd4\u5b66\u4e60\u6a21\u578b\u540e\uff0c\u4e2d\u6587\u7535\u5546\u573a\u666f\u4e0b\u7684\u56fe\u6587 \u68c0\u7d22 Recall@1 \u4ece\u7eaf\u6587\u672c\u5173\u952e\u8bcd\u5339\u914d\u7684 42% \u63d0\u5347\u5230 78%\u3002\u6587\u6863\u89e3\u6790\u573a\u666f\u4e2d\uff0c\u63a5\u5165 \u89c6\u89c9\u7f16\u7801\u5668\u540e\u626b\u63cf\u4ef6\u7684\u5b57\u6bb5\u62bd\u53d6\u51c6\u786e\u7387\u4ece 0.67 \u63d0\u5347\u5230 0.88\uff0c\u56e0\u4e3a\u8868\u683c\u548c\u5370\u7ae0 \u4fe1\u606f\u53ea\u5b58\u5728\u4e8e\u50cf\u7d20\u91cc\u3002\u97f3\u9891\u4fa7\u7684\u6536\u76ca\u7c7b\u4f3c\uff1a\u5728 6 \u5c0f\u65f6\u8bed\u97f3\u4e0e\u4e8b\u4ef6\u6587\u672c\u7684\u8054\u5408\u6570\u636e \u96c6\u4e0a\uff0c\u4e8b\u4ef6\u68c0\u6d4b F1 \u4ece 0.61 \u63d0\u5347\u5230 0.74\u3002<\/p>\n<p>\u4ece\u8fd0\u7ef4\u6210\u672c\u770b\uff0c\u4e00\u5957 7B \u591a\u6a21\u6001\u6a21\u578b\u66ff\u4ee3\u4e09\u4e2a\u5355\u6a21\u6001\u6a21\u578b\u65f6\uff0c\u63a8\u7406\u673a\u4f4d\u6570\u4ece 3 \u964d \u5230 1\uff0c\u5355\u6b21\u8bf7\u6c42\u7684\u5e73\u5747\u663e\u5b58\u5360\u7528\u4e0b\u964d\u7ea6 15%\uff0c\u56e0\u4e3a\u5355\u6a21\u6001\u6a21\u578b\u5728\u5404\u81ea\u6808\u4e2d\u91cd\u590d\u4e86 \u5927\u91cf\u5e95\u5c42\u7279\u5f81\u63d0\u53d6\u5c42\u3002\u7edf\u4e00\u8868\u793a\u4e0d\u662f\u5b66\u672f\u88c5\u9970\uff0c\u800c\u662f\u63a8\u7406\u96c6\u7fa4\u5c42\u9762\u7684\u8d44\u6e90\u8d26\u3002<\/p>\n<p>\u8fb9\u754c\u6761\u4ef6\u540c\u6837\u660e\u786e\u3002\u5f53\u8f93\u5165\u53ea\u6709\u6587\u672c\u65f6\uff0c\u591a\u6a21\u6001\u6a21\u578b\u6bd4\u540c\u89c4\u6a21\u7eaf\u6587\u672c\u6a21\u578b\u6162 8% \u5230 12%\uff0c\u6765\u81ea\u89c6\u89c9\u7f16\u7801\u5668\u7684\u989d\u5916\u524d\u5411\u5f00\u9500\uff0c\u56e0\u6b64\u7ebf\u4e0a\u8981\u6309\u8f93\u5165\u7c7b\u578b\u8def\u7531\uff0c\u7eaf\u6587\u672c \u8bf7\u6c42\u76f4\u63a5\u8d70\u7cbe\u7b80\u8def\u5f84\u3002\u53e6\u5916\uff0c\u97f3\u9891\u6a21\u6001\u5728\u4e2d\u6587\u573a\u666f\u7684\u6210\u719f\u5ea6\u4f4e\u4e8e\u56fe\u6587\uff0c\u516c\u5f00\u53ef\u7528 \u7684\u4e2d\u6587\u8bed\u97f3\u6587\u672c\u5bf9\u9f50\u6570\u636e\u5927\u7ea6\u53ea\u6709\u82f1\u6587\u7684\u5341\u5206\u4e4b\u4e00\uff0c\u6548\u679c\u4e0a\u9650\u53d7\u6570\u636e\u89c4\u6a21\u7ea6\u675f\u3002<\/p>\n<p><span class=\"token comment\"># \u6765\u6e90\uff1a\u81ea\u5b9e\u73b0 \/ dataset_stats.py<\/span><br \/>\n<span class=\"token keyword\">import<\/span> random<br \/>\n<span class=\"token keyword\">from<\/span> collections <span class=\"token keyword\">import<\/span> Counter<\/p>\n<p><span class=\"token keyword\">class<\/span> <span class=\"token class-name\">ModalityStats<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    <span class=\"token triple-quoted-string string\">\\&#8221;\\&#8221;\\&#8221;\u7edf\u8ba1\u591a\u6a21\u6001\u6570\u636e\u96c6\u4e2d\u5404\u7c7b\u6837\u672c\u7684\u6570\u91cf\u4e0e\u5360\u6bd4\u3002\\&#8221;\\&#8221;\\&#8221;<\/span><\/p>\n<p>    <span class=\"token keyword\">def<\/span> <span class=\"token function\">__init__<\/span><span class=\"token punctuation\">(<\/span>self<span class=\"token punctuation\">,<\/span> samples<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n        self<span class=\"token punctuation\">.<\/span>samples <span class=\"token operator\">=<\/span> samples  <span class=\"token comment\"># \u6bcf\u4e2a\u6837\u672c\u4e3a (\u540d\u79f0, \u5185\u5bb9, \u6a21\u6001)<\/span><br \/>\n        self<span class=\"token punctuation\">.<\/span>counter <span class=\"token operator\">=<\/span> Counter<span class=\"token punctuation\">(<\/span>m <span class=\"token keyword\">for<\/span> _<span class=\"token punctuation\">,<\/span> _<span class=\"token punctuation\">,<\/span> m <span class=\"token keyword\">in<\/span> samples<span class=\"token punctuation\">)<\/span><\/p>\n<p>    <span class=\"token keyword\">def<\/span> <span class=\"token function\">ratio<\/span><span class=\"token punctuation\">(<\/span>self<span class=\"token punctuation\">,<\/span> modality<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n        total <span class=\"token operator\">=<\/span> <span class=\"token builtin\">len<\/span><span class=\"token punctuation\">(<\/span>self<span class=\"token punctuation\">.<\/span>samples<span class=\"token punctuation\">)<\/span><br \/>\n        <span class=\"token keyword\">return<\/span> self<span class=\"token punctuation\">.<\/span>counter<span class=\"token punctuation\">.<\/span>get<span class=\"token punctuation\">(<\/span>modality<span class=\"token punctuation\">,<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">)<\/span> <span class=\"token operator\">\/<\/span> total <span class=\"token keyword\">if<\/span> total <span class=\"token keyword\">else<\/span> <span class=\"token number\">0.0<\/span><\/p>\n<p>    <span class=\"token keyword\">def<\/span> <span class=\"token function\">sample_balanced<\/span><span class=\"token punctuation\">(<\/span>self<span class=\"token punctuation\">,<\/span> modality<span class=\"token punctuation\">,<\/span> k<span class=\"token operator\">=<\/span><span class=\"token number\">8<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n        pool <span class=\"token operator\">=<\/span> <span class=\"token punctuation\">[<\/span>s <span class=\"token keyword\">for<\/span> s <span class=\"token keyword\">in<\/span> self<span class=\"token punctuation\">.<\/span>samples <span class=\"token keyword\">if<\/span> s<span class=\"token punctuation\">[<\/span><span class=\"token number\">2<\/span><span class=\"token punctuation\">]<\/span> <span class=\"token operator\">==<\/span> modality<span class=\"token punctuation\">]<\/span><br \/>\n        <span class=\"token keyword\">return<\/span> random<span class=\"token punctuation\">.<\/span>sample<span class=\"token punctuation\">(<\/span>pool<span class=\"token punctuation\">,<\/span> <span class=\"token builtin\">min<\/span><span class=\"token punctuation\">(<\/span>k<span class=\"token punctuation\">,<\/span> <span class=\"token builtin\">len<\/span><span class=\"token punctuation\">(<\/span>pool<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><\/p>\n<p><span class=\"token keyword\">if<\/span> __name__ <span class=\"token operator\">==<\/span> <span class=\"token string\">\\&#8221;__main__\\&#8221;<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    data <span class=\"token operator\">=<\/span> <span class=\"token punctuation\">[<\/span><br \/>\n        <span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;doc_0001.txt\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;\u670d\u52a1\u6761\u6b3e\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;\u6587\u672c\\&#8221;<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n        <span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;img_0001.jpg\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;\u7a97\u53f0\u4e0a\u7684\u732b\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;\u56fe\u50cf\\&#8221;<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n        <span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;clip_0001.mp3\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;\u96e8\u58f0\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;\u97f3\u9891\\&#8221;<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n    <span class=\"token punctuation\">]<\/span> <span class=\"token operator\">*<\/span> <span class=\"token number\">10<\/span><br \/>\n    stats <span class=\"token operator\">=<\/span> ModalityStats<span class=\"token punctuation\">(<\/span>data<span class=\"token punctuation\">)<\/span><br \/>\n    <span class=\"token keyword\">for<\/span> m <span class=\"token keyword\">in<\/span> <span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;\u6587\u672c\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;\u56fe\u50cf\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;\u97f3\u9891\\&#8221;<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n        <span class=\"token keyword\">print<\/span><span class=\"token punctuation\">(<\/span>m<span class=\"token punctuation\">,<\/span> <span class=\"token string-interpolation\"><span class=\"token string\">f\\&#8221;<\/span><span class=\"token interpolation\"><span class=\"token punctuation\">{<\/p>\n<p>     <\/span>stats<span class=\"token punctuation\">.<\/span>ratio<span class=\"token punctuation\">(<\/span>m<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><span class=\"token format-spec\">.2%<\/span><span class=\"token punctuation\">}<\/span><\/span><span class=\"token string\">\\&#8221;<\/span><\/span><span class=\"token punctuation\">)<\/span><br \/>\n    <span class=\"token keyword\">print<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;\u91c7\u6837\u56fe\u50cf\u6570:\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token builtin\">len<\/span><span class=\"token punctuation\">(<\/span>stats<span class=\"token punctuation\">.<\/span>sample_balanced<span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;\u56fe\u50cf\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token number\">5<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><\/p>\n<h4>1.2 \u591a\u6a21\u6001\u67b6\u6784<\/h4>\n<p>\u4e3b\u6d41\u591a\u6a21\u6001\u67b6\u6784\u5206\u4e3a\u4e09\u7c7b\u3002\u62fc\u63a5\u5f0f\uff08\u65e9\u671f\u878d\u5408\uff09\u628a\u89c6\u89c9 token \u76f4\u63a5\u653e\u8fdb\u8f93\u5165\u5e8f\u5217\uff0c \u4ee3\u8868\u662f LLaVA\uff1b\u6ce8\u5165\u5f0f\uff08\u4e2d\u671f\u878d\u5408\uff09\u5728 LLM \u5c42\u95f4\u63d2\u5165\u4ea4\u53c9\u6ce8\u610f\u529b\uff0c\u4ee3\u8868\u662f Flamingo\uff1b\u539f\u751f\u7edf\u4e00\u5f0f\u628a\u56fe\u50cf\u79bb\u6563\u6210 token \u4e0e\u6587\u672c\u8d70\u540c\u4e00\u81ea\u56de\u5f52\u8def\u5f84\uff0c\u4ee3\u8868\u662f Chameleon\u3002\u4e09\u8005\u7684\u5dee\u5f02\u96c6\u4e2d\u5728\u89c6\u89c9\u4fe1\u606f\u5728\u4f55\u65f6\u4ee5\u4f55\u79cd\u5f62\u5f0f\u8fdb\u5165 Transformer\u3002<\/p>\n<p>\u62fc\u63a5\u5f0f\u5b9e\u73b0\u6700\u7b80\u5355\u3002LLaVA-1.5 \u7528 336 \u4e58 336 \u8f93\u5165\u300114 \u4e58 14 \u5206\u5757\uff0c\u4ea7\u751f 576 \u4e2a\u89c6\u89c9 token\uff0c\u5360 2048 \u957f\u5ea6\u4e0a\u4e0b\u6587\u7684 28%\u3002\u89c6\u89c9 token \u4e0e\u6587\u672c\u5171\u4eab\u5168\u90e8\u540e\u7eed\u5c42\uff0c \u63a8\u7406\u65f6\u53ea\u9700\u8981\u4e00\u6b21\u524d\u5411\uff0c\u4f46\u4e0a\u4e0b\u6587\u91cc\u89c6\u89c9\u5360\u6bd4\u8fc7\u9ad8\u65f6\uff0c\u957f\u6587\u672c\u63a8\u7406\u80fd\u529b\u88ab\u538b\u7f29\u3002<\/p>\n<p>\u6ce8\u5165\u5f0f\u4e0d\u5360\u8f93\u5165\u957f\u5ea6\u3002Flamingo \u7528 Perceiver Resampler \u628a\u89c6\u9891\u5e27\u4ea7\u751f\u7684\u6570\u4e07 \u89c6\u89c9 token \u538b\u7f29\u6210\u56fa\u5b9a 64 \u4e2a\uff0c\u518d\u901a\u8fc7\u95e8\u63a7\u4ea4\u53c9\u6ce8\u610f\u529b\u6ce8\u5165\uff0c\u95e8\u63a7\u53c2\u6570\u521d\u59cb\u5316\u4e3a 0\uff0c\u4fdd\u8bc1\u8bad\u7ec3\u65e9\u671f\u89c6\u89c9\u4fe1\u606f\u4e0d\u4f1a\u7834\u574f\u5df2\u9884\u8bad\u7ec3\u7684\u6587\u672c\u6743\u91cd\u3002\u8fd9\u4e2a\u95e8\u63a7\u4ece 0 \u8d77\u6b65\u7684 \u521d\u59cb\u5316\u662f\u591a\u6a21\u6001\u8bad\u7ec3\u7a33\u5b9a\u6027\u7684\u5173\u952e\u5de5\u7a0b\u7ec6\u8282\u3002<\/p>\n<p>\u7edf\u4e00\u5f0f\u67b6\u6784\u6700\u5e72\u51c0\u3002\u56fe\u50cf patch \u7ecf VQ-VAE \u91cf\u5316\u6210\u79bb\u6563\u7801\uff0c\u4e0e\u6587\u672c token \u5171\u7528 \u8bcd\u8868\u4e0e\u89e3\u7801\u8def\u5f84\uff0c\u4f46 8 \u4e07\u8bcd\u6c47\u8868\u4e2d\u8981\u4e3a\u56fe\u50cf\u7801\u9884\u7559\u7ea6 1 \u4e07\u4e2a\u69fd\u4f4d\uff0c\u4e14\u91cf\u5316\u7f16\u7801\u5668 \u9700\u8981\u72ec\u7acb\u8bad\u7ec3\u3002\u4e09\u7c7b\u67b6\u6784\u5728\u878d\u5408\u5c42\u7684\u53c2\u6570\u91cf\u5dee\u5f02\u660e\u663e\uff1a\u62fc\u63a5\u5f0f\u589e\u52a0\u4e00\u4e2a\u6295\u5f71\u5668 \uff084M \u5230 21M\uff09\uff0c\u6ce8\u5165\u5f0f\u589e\u52a0 Resampler \u4e0e\u4ea4\u53c9\u6ce8\u610f\u529b\u6a21\u5757\uff08\u7ea6 500M\uff09\uff0c\u7edf\u4e00\u5f0f \u589e\u52a0 VQ \u89e3\u7801\u5668\uff08\u7ea6 300M\uff09\u3002<\/p>\n<p>\u6f14\u8fdb\u8d8b\u52bf\u504f\u5411\u7edf\u4e00\u5f0f\uff0c\u516c\u5f00\u6280\u672f\u62a5\u544a\u90fd\u503e\u5411\u4e00\u5207\u8f93\u5165\u8fdb\u540c\u4e00\u4e2a Transformer\uff0c\u4f46 \u62fc\u63a5\u5f0f\u4ecd\u662f\u5f00\u6e90\u751f\u6001\u91cc\u6027\u4ef7\u6bd4\u6700\u9ad8\u7684\u57fa\u7ebf\uff0c\u5176\u4ee3\u7801\u4e0e\u6570\u636e\u7ba1\u7ebf\u6700\u6210\u719f\uff0c\u51b7\u542f\u52a8\u6210\u672c \u6700\u4f4e\u3002\u4e09\u7c7b\u67b6\u6784\u7684\u5bf9\u6bd4\u89c1\u8868 1-1\u3002<\/p>\n<table>\n<tr>\n   \u67b6\u6784\u7c7b\u578b<br \/>\n   \u4ee3\u8868\u6a21\u578b<br \/>\n   \u878d\u5408\u4f4d\u7f6e<br \/>\n   \u89c6\u89c9 token \u6570\u91cf<br \/>\n   \u878d\u5408\u5c42\u53c2\u6570<br \/>\n  <\/tr>\n<tbody>\n<tr>\n<td>\u62fc\u63a5\u5f0f<\/td>\n<td>LLaVA-1.5<\/td>\n<td>\u8f93\u5165\u5e8f\u5217<\/td>\n<td>576<\/td>\n<td>4M \u5230 21M<\/td>\n<\/tr>\n<tr>\n<td>\u6ce8\u5165\u5f0f<\/td>\n<td>Flamingo<\/td>\n<td>\u5c42\u95f4\u4ea4\u53c9\u6ce8\u610f\u529b<\/td>\n<td>64 \u56fa\u5b9a<\/td>\n<td>\u7ea6 500M<\/td>\n<\/tr>\n<tr>\n<td>\u7edf\u4e00\u5f0f<\/td>\n<td>Chameleon<\/td>\n<td>\u5171\u4eab\u8bcd\u8868<\/td>\n<td>\u79bb\u6563\u7801<\/td>\n<td>\u7ea6 300M<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u8868 1-1 \u4e09\u79cd\u591a\u6a21\u6001\u67b6\u6784\u7684\u878d\u5408\u65b9\u5f0f\u4e0e\u5f00\u9500\u5bf9\u6bd4\u3002<\/p>\n<p><span class=\"token comment\"># \u6765\u6e90\uff1a\u81ea\u5b9e\u73b0 \/ registry.py<\/span><br \/>\n<span class=\"token keyword\">class<\/span> <span class=\"token class-name\">EncoderRegistry<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    <span class=\"token triple-quoted-string string\">\\&#8221;\\&#8221;\\&#8221;\u6309\u540d\u79f0\u6784\u9020\u89c6\u89c9\u7f16\u7801\u5668\uff0c\u907f\u514d\u5728\u8bad\u7ec3\u811a\u672c\u91cc\u5806\u786c\u7f16\u7801\u5206\u652f\u3002\\&#8221;\\&#8221;\\&#8221;<\/span><\/p>\n<p>    _registry <span class=\"token operator\">=<\/span> <span class=\"token punctuation\">{<\/p>\n<p>   <\/span><span class=\"token punctuation\">}<\/span><\/p>\n<p>    <span class=\"token decorator annotation punctuation\">@classmethod<\/span><br \/>\n    <span class=\"token keyword\">def<\/span> <span class=\"token function\">register<\/span><span class=\"token punctuation\">(<\/span>cls<span class=\"token punctuation\">,<\/span> name<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n        <span class=\"token keyword\">def<\/span> <span class=\"token function\">deco<\/span><span class=\"token punctuation\">(<\/span>factory<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n            cls<span class=\"token punctuation\">.<\/span>_registry<span class=\"token punctuation\">[<\/span>name<span class=\"token punctuation\">]<\/span> <span class=\"token operator\">=<\/span> factory<br \/>\n            <span class=\"token keyword\">return<\/span> factory<br \/>\n        <span class=\"token keyword\">return<\/span> deco<\/p>\n<p>    <span class=\"token decorator annotation punctuation\">@classmethod<\/span><br \/>\n    <span class=\"token keyword\">def<\/span> <span class=\"token function\">build<\/span><span class=\"token punctuation\">(<\/span>cls<span class=\"token punctuation\">,<\/span> name<span class=\"token punctuation\">,<\/span> <span class=\"token operator\">*<\/span>args<span class=\"token punctuation\">,<\/span> <span class=\"token operator\">**<\/span>kwargs<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n        <span class=\"token keyword\">if<\/span> name <span class=\"token keyword\">not<\/span> <span class=\"token keyword\">in<\/span> cls<span class=\"token punctuation\">.<\/span>_registry<span class=\"token punctuation\">:<\/span><br \/>\n            <span class=\"token keyword\">raise<\/span> KeyError<span class=\"token punctuation\">(<\/span><span class=\"token string-interpolation\"><span class=\"token string\">f\\&#8221;\u672a\u6ce8\u518c\u7684\u7f16\u7801\u5668: <\/span><span class=\"token interpolation\"><span class=\"token punctuation\">{<\/p>\n<p>     <\/span>name<span class=\"token punctuation\">}<\/span><\/span><span class=\"token string\">\\&#8221;<\/span><\/span><span class=\"token punctuation\">)<\/span><br \/>\n        <span class=\"token keyword\">return<\/span> cls<span class=\"token punctuation\">.<\/span>_registry<span class=\"token punctuation\">[<\/span>name<span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">*<\/span>args<span class=\"token punctuation\">,<\/span> <span class=\"token operator\">**<\/span>kwargs<span class=\"token punctuation\">)<\/span><\/p>\n<p><span class=\"token decorator annotation punctuation\">@EncoderRegistry<span class=\"token punctuation\">.<\/span>register<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;vit_b_16\\&#8221;<\/span><span class=\"token punctuation\">)<\/span><br \/>\n<span class=\"token keyword\">def<\/span> <span class=\"token function\">build_vit_b<\/span><span class=\"token punctuation\">(<\/span>hidden<span class=\"token operator\">=<\/span><span class=\"token number\">768<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    <span class=\"token keyword\">return<\/span> <span class=\"token punctuation\">{<\/p>\n<p>   <\/span><span class=\"token string\">\\&#8221;type\\&#8221;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token string\">\\&#8221;ViT-B\/16\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;hidden\\&#8221;<\/span><span class=\"token punctuation\">:<\/span> hidden<span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;patches\\&#8221;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token number\">196<\/span><span class=\"token punctuation\">}<\/span><\/p>\n<p><span class=\"token decorator annotation punctuation\">@EncoderRegistry<span class=\"token punctuation\">.<\/span>register<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;clip_l_14\\&#8221;<\/span><span class=\"token punctuation\">)<\/span><br \/>\n<span class=\"token keyword\">def<\/span> <span class=\"token function\">build_clip_l<\/span><span class=\"token punctuation\">(<\/span>hidden<span class=\"token operator\">=<\/span><span class=\"token number\">1024<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    <span class=\"token keyword\">return<\/span> <span class=\"token punctuation\">{<\/p>\n<p>   <\/span><span class=\"token string\">\\&#8221;type\\&#8221;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token string\">\\&#8221;CLIP-L\/14\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;hidden\\&#8221;<\/span><span class=\"token punctuation\">:<\/span> hidden<span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;patches\\&#8221;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token number\">256<\/span><span class=\"token punctuation\">}<\/span><\/p>\n<p><span class=\"token keyword\">if<\/span> __name__ <span class=\"token operator\">==<\/span> <span class=\"token string\">\\&#8221;__main__\\&#8221;<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    <span class=\"token keyword\">for<\/span> name <span class=\"token keyword\">in<\/span> <span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;vit_b_16\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;clip_l_14\\&#8221;<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n        enc <span class=\"token operator\">=<\/span> EncoderRegistry<span class=\"token punctuation\">.<\/span>build<span class=\"token punctuation\">(<\/span>name<span class=\"token punctuation\">)<\/span><br \/>\n        <span class=\"token keyword\">print<\/span><span class=\"token punctuation\">(<\/span>name<span class=\"token punctuation\">,<\/span> enc<span class=\"token punctuation\">[<\/span><span class=\"token string\">\\&#8221;type\\&#8221;<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;hidden=\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> enc<span class=\"token punctuation\">[<\/span><span class=\"token string\">\\&#8221;hidden\\&#8221;<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">)<\/span><\/p>\n<h4>1.3 \u591a\u6a21\u6001\u6570\u636e\u7684\u6311\u6218<\/h4>\n<p>\u591a\u6a21\u6001\u6570\u636e\u7684\u4e3b\u8981\u95ee\u9898\u4f9d\u6b21\u662f\u56fe\u6587\u5bf9\u566a\u58f0\u3001\u6807\u6ce8\u7a00\u758f\u3001\u6a21\u6001\u5c3a\u5ea6\u4e0d\u4e00\u81f4\u3002<\/p>\n<p>\u56fe\u6587\u5bf9\u566a\u58f0\u6307\u56fe\u50cf\u4e0e\u6587\u672c\u63cf\u8ff0\u4e0d\u5339\u914d\u3002LAION-5B \u4e2d\u4f30\u8ba1\u6709 15% \u5230 30% \u7684\u566a\u58f0\u5bf9\uff0c \u76f4\u63a5\u8bad\u7ec3\u4f1a\u62c9\u4f4e\u56fe\u6587\u68c0\u7d22\u51c6\u786e\u7387\u7ea6 5 \u4e2a\u767e\u5206\u70b9\u3002\u5b9e\u8df5\u505a\u6cd5\u662f\u7528 CLIP \u76f8\u4f3c\u5ea6\u8fc7\u6ee4\uff0c \u5f52\u4e00\u5316\u4f59\u5f26\u4f4e\u4e8e 0.25 \u7684\u6837\u672c\u4e22\u5f03\uff0c\u80fd\u56de\u6536\u7ea6 25% \u7684\u65e0\u6548\u7b97\u529b\uff0c\u4f46\u9608\u503c\u8c03\u9ad8\u4f1a\u8bef \u6740\u957f\u5c3e\u4e3b\u9898\uff0c\u8d85\u8fc7 0.30 \u65f6\u53ec\u56de\u635f\u5931\u5f00\u59cb\u5927\u4e8e\u566a\u58f0\u6536\u76ca\u3002<\/p>\n<p>\u6807\u6ce8\u7a00\u758f\u96c6\u4e2d\u5728\u89c6\u89c9\u5b9a\u4f4d\u4e0e\u533a\u57df\u63cf\u8ff0\u3002COCO \u53ea\u6709\u7ea6 33 \u4e07\u5f20\u5e26\u5b9e\u4f8b\u6807\u6ce8\u7684\u56fe\u50cf\uff0c \u8fdc\u5c0f\u4e8e\u7f51\u9875\u6293\u53d6\u7684\u4ebf\u7ea7\u89c4\u6a21\u3002\u5bf9\u7b56\u662f\u8bfe\u7a0b\u5f0f\u914d\u6bd4\uff1a\u5148\u7528\u51e0\u5341\u4e07\u6761\u9ad8\u8d28\u91cf\u6807\u6ce8\u786e\u5b9a \u5bf9\u9f50\u8d28\u91cf\uff0c\u518d\u5e76\u5165\u4ebf\u7ea7\u5f31\u6807\u6ce8\u5bf9\u6269\u5145\u8986\u76d6\uff0c\u5f31\u6807\u6ce8\u4e0e\u5f3a\u6807\u6ce8\u7684\u914d\u6bd4\u901a\u5e38\u5728 5 \u6bd4 1 \u5230 10 \u6bd4 1\u3002<\/p>\n<p>\u6a21\u6001\u5c3a\u5ea6\u4e0d\u4e00\u81f4\u4f53\u73b0\u5728 token \u7c92\u5ea6\uff1a\u4e2d\u6587\u6587\u672c\u7ea6 0.7 token \u6bcf\u5b57\uff0c\u56fe\u50cf\u6309 576 token \u6bcf\u5f20\u6298\u7b97\uff0c\u97f3\u9891\u7ea6 100 token \u6bcf\u79d2\u3002\u4e09\u8005\u8fdb\u5165\u540c\u4e00\u4e2a\u4e0a\u4e0b\u6587\u7a97\u53e3\u65f6\uff0c \u9700\u8981\u6309\u8ba1\u7b97\u9884\u7b97\u6362\u7b97\u5360\u6bd4\uff0c\u5426\u5219\u67d0\u7c7b\u6a21\u6001\u4f1a\u5784\u65ad\u4e0a\u4e0b\u6587\uff0c\u538b\u5236\u5176\u4ed6\u6a21\u6001\u7684\u8868\u8fbe\u3002<\/p>\n<p>\u5408\u89c4\u7ea6\u675f\u4f1a\u6539\u53d8\u6570\u636e\u7ba1\u7ebf\u3002\u6293\u53d6\u6570\u636e\u9700\u505a\u4eba\u8138\u68c0\u6d4b\uff08\u53ec\u56de\u7387\u7ea6 0.98\uff09\u4e0e\u6587\u5b57\u6c34\u5370 \u8fc7\u6ee4\uff0c\u8fd9\u4e00\u6b65\u4f7f\u6570\u636e\u7ba1\u7ebf\u541e\u5410\u4e0b\u964d\u7ea6 20%\uff0c\u9700\u8981\u5728\u9879\u76ee\u6392\u671f\u91cc\u9884\u7559\u3002\u6570\u636e\u95ee\u9898\u7684 \u6574\u4f53\u4f18\u5148\u7ea7\u662f\uff1a\u5148\u89e3\u51b3\u566a\u58f0\uff0c\u518d\u89e3\u51b3\u7a00\u758f\uff0c\u6700\u540e\u624d\u8c08\u89c4\u6a21\u3002<\/p>\n<p><span class=\"token comment\"># \u6765\u6e90\uff1a\u81ea\u5b9e\u73b0 \/ filter_pairs.py<\/span><br \/>\n<span class=\"token keyword\">import<\/span> hashlib<\/p>\n<p><span class=\"token keyword\">def<\/span> <span class=\"token function\">dedup_and_filter<\/span><span class=\"token punctuation\">(<\/span>pairs<span class=\"token punctuation\">,<\/span> min_chars<span class=\"token operator\">=<\/span><span class=\"token number\">8<\/span><span class=\"token punctuation\">,<\/span> max_ratio<span class=\"token operator\">=<\/span><span class=\"token number\">0.95<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    <span class=\"token triple-quoted-string string\">\\&#8221;\\&#8221;\\&#8221;\u53bb\u9664\u91cd\u590d\u56fe\u6587\u5bf9\uff0c\u8fc7\u6ee4\u6587\u672c\u8fc7\u77ed\u6216\u63cf\u8ff0\u6bd4\u4f8b\u5f02\u5e38\u7684\u6837\u672c\u3002\\&#8221;\\&#8221;\\&#8221;<\/span><br \/>\n    seen <span class=\"token operator\">=<\/span> <span class=\"token builtin\">set<\/span><span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">)<\/span><br \/>\n    kept <span class=\"token operator\">=<\/span> <span class=\"token punctuation\">[<\/span><span class=\"token punctuation\">]<\/span><br \/>\n    <span class=\"token keyword\">for<\/span> image_path<span class=\"token punctuation\">,<\/span> caption <span class=\"token keyword\">in<\/span> pairs<span class=\"token punctuation\">:<\/span><br \/>\n        digest <span class=\"token operator\">=<\/span> hashlib<span class=\"token punctuation\">.<\/span>md5<span class=\"token punctuation\">(<\/span>image_path<span class=\"token punctuation\">.<\/span>encode<span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;utf-8\\&#8221;<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">.<\/span>hexdigest<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">)<\/span><br \/>\n        <span class=\"token keyword\">if<\/span> digest <span class=\"token keyword\">in<\/span> seen<span class=\"token punctuation\">:<\/span><br \/>\n            <span class=\"token keyword\">continue<\/span><br \/>\n        <span class=\"token keyword\">if<\/span> <span class=\"token builtin\">len<\/span><span class=\"token punctuation\">(<\/span>caption<span class=\"token punctuation\">.<\/span>strip<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span> <span class=\"token operator\">&lt;<\/span> min_chars<span class=\"token punctuation\">:<\/span><br \/>\n            <span class=\"token keyword\">continue<\/span><br \/>\n        <span class=\"token comment\"># \u4ece\u6587\u4ef6\u540d\u63d0\u53d6\u56fe\u50cf\u89c4\u6a21\u505a\u542f\u53d1\u5f0f\u8fc7\u6ee4\uff0c\u5254\u9664\u566a\u58f0\u5bf9<\/span><br \/>\n        <span class=\"token keyword\">try<\/span><span class=\"token punctuation\">:<\/span><br \/>\n            size <span class=\"token operator\">=<\/span> <span class=\"token builtin\">int<\/span><span class=\"token punctuation\">(<\/span>image_path<span class=\"token punctuation\">.<\/span>rsplit<span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;_\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token number\">1<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">[<\/span><span class=\"token number\">1<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">.<\/span>split<span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;.\\&#8221;<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">[<\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">)<\/span><br \/>\n        <span class=\"token keyword\">except<\/span> <span class=\"token punctuation\">(<\/span>ValueError<span class=\"token punctuation\">,<\/span> IndexError<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n            size <span class=\"token operator\">=<\/span> <span class=\"token number\">1<\/span><br \/>\n        <span class=\"token keyword\">if<\/span> size <span class=\"token keyword\">and<\/span> <span class=\"token builtin\">len<\/span><span class=\"token punctuation\">(<\/span>caption<span class=\"token punctuation\">)<\/span> <span class=\"token operator\">\/<\/span> size <span class=\"token operator\">&gt;<\/span> max_ratio<span class=\"token punctuation\">:<\/span><br \/>\n            <span class=\"token keyword\">continue<\/span><br \/>\n        seen<span class=\"token punctuation\">.<\/span>add<span class=\"token punctuation\">(<\/span>digest<span class=\"token punctuation\">)<\/span><br \/>\n        kept<span class=\"token punctuation\">.<\/span>append<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">(<\/span>image_path<span class=\"token punctuation\">,<\/span> caption<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><br \/>\n    <span class=\"token keyword\">return<\/span> kept<\/p>\n<p><span class=\"token keyword\">if<\/span> __name__ <span class=\"token operator\">==<\/span> <span class=\"token string\">\\&#8221;__main__\\&#8221;<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    raw <span class=\"token operator\">=<\/span> <span class=\"token punctuation\">[<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;a_1200.jpg\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;\u96ea\u5c71\u4e0a\u7684\u65e5\u51fa\\&#8221;<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">]<\/span> <span class=\"token operator\">*<\/span> <span class=\"token number\">3<\/span> <span class=\"token operator\">+<\/span> <span class=\"token punctuation\">[<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">\\&#8221;b_5.jpg\\&#8221;<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\\&#8221;hi\\&#8221;<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">]<\/span><br \/>\n    <span class=\"token keyword\">print<\/span><span class=\"token punctuation\">(<\/span><span class=\"token builtin\">len<\/span><span class=\"token punctuation\">(<\/span>dedup_and_filter<span class=\"token punctuation\">(<\/span>raw<span class=\"token punctuation\">,<\/span> min_chars<span class=\"token operator\">=<\/span><span class=\"token number\">4<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><\/p>\n<h3>2. \u89c6\u89c9\u7f16\u7801\u5668\uff1aViT \u4e0e CLIP<\/h3>\n<p>\u89c6\u89c9\u7f16\u7801\u5668\u628a\u8fde\u7eed\u50cf\u7d20\u8f6c\u6362\u4e3a\u6a21\u578b\u53ef\u5904\u7406\u7684\u5411\u91cf\u5e8f\u5217\u3002\u672c\u7ae0\u5148\u62c6 ViT \u7684\u5206\u5757\u3001 \u4f4d\u7f6e\u7f16\u7801\u4e0e\u590d\u6742\u5ea6\uff0c\u518d\u8bb2 CLIP \u7684\u5bf9\u6bd4\u9884\u8bad\u7ec3\u4e0e\u56fa\u5b9a\u7528\u6cd5\uff0c\u6700\u540e\u843d\u5230\u628a\u89c6\u89c9\u7279\u5f81 \u6620\u5c04\u8fdb LLM \u8bcd\u5d4c\u5165\u7a7a\u95f4\u7684\u6295\u5f71\u5c42\u3002<\/p>\n<p>  #mermaid-svg-qd1B2e7RisyTNl93{font-family:\\&#8221;trebuchet ms\\&#8221;,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-qd1B2e7RisyTNl93 .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-qd1B2e7RisyTNl93 .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-qd1B2e7RisyTNl93 .error-icon{fill:#552222;}#mermaid-svg-qd1B2e7RisyTNl93 .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-qd1B2e7RisyTNl93 .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-qd1B2e7RisyTNl93 .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-qd1B2e7RisyTNl93 .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-qd1B2e7RisyTNl93 .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-qd1B2e7RisyTNl93 .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-qd1B2e7RisyTNl93 .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-qd1B2e7RisyTNl93 .marker{fill:#333333;stroke:#333333;}#mermaid-svg-qd1B2e7RisyTNl93 .marker.cross{stroke:#333333;}#mermaid-svg-qd1B2e7RisyTNl93 svg{font-family:\\&#8221;trebuchet ms\\&#8221;,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-qd1B2e7RisyTNl93 p{margin:0;}#mermaid-svg-qd1B2e7RisyTNl93 .label{font-family:\\&#8221;trebuchet ms\\&#8221;,verdana,arial,sans-serif;color:#333;}#mermaid-svg-qd1B2e7RisyTNl93 .cluster-label text{fill:#333;}#mermaid-svg-qd1B2e7RisyTNl93 .cluster-label span{color:#333;}#mermaid-svg-qd1B2e7RisyTNl93 .cluster-label span p{background-color:transparent;}#mermaid-svg-qd1B2e7RisyTNl93 .label text,#mermaid-svg-qd1B2e7RisyTNl93 span{fill:#333;color:#333;}#mermaid-svg-qd1B2e7RisyTNl93 .node rect,#mermaid-svg-qd1B2e7RisyTNl93 .node circle,#mermaid-svg-qd1B2e7RisyTNl93 .node ellipse,#mermaid-svg-qd1B2e7RisyTNl93 .node polygon,#mermaid-svg-qd1B2e7RisyTNl93 .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-qd1B2e7RisyTNl93 .rough-node .label text,#mermaid-svg-qd1B2e7RisyTNl93 .node .label text,#mermaid-svg-qd1B2e7RisyTNl93 .image-shape .label,#mermaid-svg-qd1B2e7RisyTNl93 .icon-shape .label{text-anchor:middle;}#mermaid-svg-qd1B2e7RisyTNl93 .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-qd1B2e7RisyTNl93 .rough-node .label,#mermaid-svg-qd1B2e7RisyTNl93 .node .label,#mermaid-svg-qd1B2e7RisyTNl93 .image-shape .label,#mermaid-svg-qd1B2e7RisyTNl93 .icon-shape .label{text-align:center;}#mermaid-svg-qd1B2e7RisyTNl93 .node.clickable{cursor:pointer;}#mermaid-svg-qd1B2e7RisyTNl93 .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-qd1B2e7RisyTNl93 .arrowheadPath{fill:#333333;}#mermaid-svg-qd1B2e7RisyTNl93 .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-qd1B2e7RisyTNl93 .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-qd1B2e7RisyTNl93 .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-qd1B2e7RisyTNl93 .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-qd1B2e7RisyTNl93 .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-qd1B2e7RisyTNl93 .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-qd1B2e7RisyTNl93 .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-qd1B2e7RisyTNl93 .cluster text{fill:#333;}#mermaid-svg-qd1B2e7RisyTNl93 .cluster span{color:#333;}#mermaid-svg-qd1B2e7RisyTNl93 div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\\&#8221;trebuchet ms\\&#8221;,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-qd1B2e7RisyTNl93 .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-qd1B2e7RisyTNl93 rect.text{fill:none;stroke-width:0;}#mermaid-svg-qd1B2e7RisyTNl93 .icon-shape,#mermaid-svg-qd1B2e7RisyTNl93 .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-qd1B2e7RisyTNl93 .icon-shape p,#mermaid-svg-qd1B2e7RisyTNl93 .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-qd1B2e7RisyTNl93 .icon-shape .label rect,#mermaid-svg-qd1B2e7RisyTNl93 .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-qd1B2e7RisyTNl93 .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-qd1B2e7RisyTNl93 .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-qd1B2e7RisyTNl93 :root{&#8211;mermaid-font-family:\\&#8221;trebuchet ms\\&#8221;,verdana,arial,sans-serif;}#mermaid-svg-qd1B2e7RisyTNl93 .default&gt;*{fill:#faf9f5!important;stroke:#ffffff!important;color:#000000!important;stroke-width:0px!important;}#mermaid-svg-qd1B2e7RisyTNl93 .default span{fill:#faf9f5!important;stroke:#ffffff!important;color:#000000!important;stroke-width:0px!important;}#mermaid-svg-qd1B2e7RisyTNl93 .default tspan{fill:#000000!important;} <\/p>\n<p>         <span class=\"nodeLabel\"><\/p>\n<p>\u6587\u672c\u7f16\u7801\u8def\u5f84<\/p>\n<p><\/span> <\/p>\n<p>         <span class=\"nodeLabel\"><\/p>\n<p>ViT \u7f16\u7801\u8def\u5f84<\/p>\n<p><\/span> <\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u76ee\u5f55 \u591a\u6a21\u6001\u8bad\u7ec3\u6982\u8ff0 \u89c6\u89c9\u7f16\u7801\u5668\uff1aViT \u4e0e CLIP \u591a\u6a21\u6001\u5bf9\u9f50\uff1a\u5bf9\u6bd4\u5b66\u4e60\u4e0e\u8de8\u6a21\u6001\u6ce8\u610f\u529b \u591a\u6a21\u6001\u878d\u5408\uff1a\u65e9\u671f\u878d\u5408\u4e0e\u665a\u671f\u878d\u5408 \u591a\u6a21\u6001\u751f\u6210\uff1a\u6587\u672c\u5230\u56fe\u50cf\u3001\u56fe\u50cf\u5230\u6587\u672c \u591a\u6a21\u6001\u8bad\u7ec3\u6311\u6218 \u5b9e\u8df5\u5efa\u8bae\u4e0e\u6700\u4f73\u5b9e\u8df5 \u6458\u8981 \u591a\u6a21\u6001\u8bad\u7ec3\u628a\u6587\u672c\u3001\u56fe\u50cf\u3001\u97f3\u9891\u6295\u5f71\u5230\u540c\u4e00\u5411\u91cf\u7a7a\u95f4\uff0c\u8ba9\u4e00\u4e2a 7B \u89c4\u6a21 LLM \u540c\u65f6\u627f\u62c5\u7406\u89e3\u4e0e\u751f\u6210\u4e24\u7c7b\u4efb\u52a1\u3002\u4ee5 CLIP-L \u89c6\u89c9\u7f16\u7801\u5668\u8f93\u51fa 1024 \u7ef4\u7279\u5f81\u3001\u5355\u5c42\u7ebf\u6027\u6295\u5f71\u6620\u5c04\u5230 4096 \u7ef4\u8bcd\u5d4c\u5165\u7684\u914d\u7f6e\u4e3a\u4f8b\uff0c\u6295\u5f71\u4ec5\u589e\u52a0\u7ea6 4.2M \u53c2\u6570\uff0c\u8bad\u7ec3\u663e\u5b58\u76f8\u6bd4\u7eaf\u6587\u672c\u591a 30% \u5230 50%\u3002<\/p>\n","protected":false},"author":2,"featured_media":92883,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[10627,10625,10626,50,956],"topic":[],"class_list":["post-92884","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-vit","tag-10625","tag-10626","tag-50","tag-956"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u591a\u6a21\u6001\u8bad\u7ec3\uff1a\u4ece\u50cf\u7d20\u5230\u8bed\u4e49\uff0c\u4e00\u4e2a\u6a21\u578b\u7edf\u6cbb\u6240\u6709\u6a21\u6001 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/92884.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u591a\u6a21\u6001\u8bad\u7ec3\uff1a\u4ece\u50cf\u7d20\u5230\u8bed\u4e49\uff0c\u4e00\u4e2a\u6a21\u578b\u7edf\u6cbb\u6240\u6709\u6a21\u6001 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u76ee\u5f55 \u591a\u6a21\u6001\u8bad\u7ec3\u6982\u8ff0 \u89c6\u89c9\u7f16\u7801\u5668\uff1aViT \u4e0e CLIP \u591a\u6a21\u6001\u5bf9\u9f50\uff1a\u5bf9\u6bd4\u5b66\u4e60\u4e0e\u8de8\u6a21\u6001\u6ce8\u610f\u529b \u591a\u6a21\u6001\u878d\u5408\uff1a\u65e9\u671f\u878d\u5408\u4e0e\u665a\u671f\u878d\u5408 \u591a\u6a21\u6001\u751f\u6210\uff1a\u6587\u672c\u5230\u56fe\u50cf\u3001\u56fe\u50cf\u5230\u6587\u672c \u591a\u6a21\u6001\u8bad\u7ec3\u6311\u6218 \u5b9e\u8df5\u5efa\u8bae\u4e0e\u6700\u4f73\u5b9e\u8df5 \u6458\u8981 \u591a\u6a21\u6001\u8bad\u7ec3\u628a\u6587\u672c\u3001\u56fe\u50cf\u3001\u97f3\u9891\u6295\u5f71\u5230\u540c\u4e00\u5411\u91cf\u7a7a\u95f4\uff0c\u8ba9\u4e00\u4e2a 7B \u89c4\u6a21 LLM \u540c\u65f6\u627f\u62c5\u7406\u89e3\u4e0e\u751f\u6210\u4e24\u7c7b\u4efb\u52a1\u3002\u4ee5 CLIP-L \u89c6\u89c9\u7f16\u7801\u5668\u8f93\u51fa 1024 \u7ef4\u7279\u5f81\u3001\u5355\u5c42\u7ebf\u6027\u6295\u5f71\u6620\u5c04\u5230 4096 \u7ef4\u8bcd\u5d4c\u5165\u7684\u914d\u7f6e\u4e3a\u4f8b\uff0c\u6295\u5f71\u4ec5\u589e\u52a0\u7ea6 4.2M \u53c2\u6570\uff0c\u8bad\u7ec3\u663e\u5b58\u76f8\u6bd4\u7eaf\u6587\u672c\u591a 30% \u5230 50%\u3002\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/92884.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-08-11T05:56:48+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260811055646-6a7ab99ee25b3.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"7 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/92884.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/92884.html\",\"name\":\"\u591a\u6a21\u6001\u8bad\u7ec3\uff1a\u4ece\u50cf\u7d20\u5230\u8bed\u4e49\uff0c\u4e00\u4e2a\u6a21\u578b\u7edf\u6cbb\u6240\u6709\u6a21\u6001 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-08-11T05:56:48+00:00\",\"dateModified\":\"2026-08-11T05:56:48+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/92884.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/92884.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/92884.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u591a\u6a21\u6001\u8bad\u7ec3\uff1a\u4ece\u50cf\u7d20\u5230\u8bed\u4e49\uff0c\u4e00\u4e2a\u6a21\u578b\u7edf\u6cbb\u6240\u6709\u6a21\u6001\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u591a\u6a21\u6001\u8bad\u7ec3\uff1a\u4ece\u50cf\u7d20\u5230\u8bed\u4e49\uff0c\u4e00\u4e2a\u6a21\u578b\u7edf\u6cbb\u6240\u6709\u6a21\u6001 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/92884.html","og_locale":"zh_CN","og_type":"article","og_title":"\u591a\u6a21\u6001\u8bad\u7ec3\uff1a\u4ece\u50cf\u7d20\u5230\u8bed\u4e49\uff0c\u4e00\u4e2a\u6a21\u578b\u7edf\u6cbb\u6240\u6709\u6a21\u6001 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u76ee\u5f55 \u591a\u6a21\u6001\u8bad\u7ec3\u6982\u8ff0 \u89c6\u89c9\u7f16\u7801\u5668\uff1aViT \u4e0e CLIP \u591a\u6a21\u6001\u5bf9\u9f50\uff1a\u5bf9\u6bd4\u5b66\u4e60\u4e0e\u8de8\u6a21\u6001\u6ce8\u610f\u529b \u591a\u6a21\u6001\u878d\u5408\uff1a\u65e9\u671f\u878d\u5408\u4e0e\u665a\u671f\u878d\u5408 \u591a\u6a21\u6001\u751f\u6210\uff1a\u6587\u672c\u5230\u56fe\u50cf\u3001\u56fe\u50cf\u5230\u6587\u672c \u591a\u6a21\u6001\u8bad\u7ec3\u6311\u6218 \u5b9e\u8df5\u5efa\u8bae\u4e0e\u6700\u4f73\u5b9e\u8df5 \u6458\u8981 \u591a\u6a21\u6001\u8bad\u7ec3\u628a\u6587\u672c\u3001\u56fe\u50cf\u3001\u97f3\u9891\u6295\u5f71\u5230\u540c\u4e00\u5411\u91cf\u7a7a\u95f4\uff0c\u8ba9\u4e00\u4e2a 7B \u89c4\u6a21 LLM \u540c\u65f6\u627f\u62c5\u7406\u89e3\u4e0e\u751f\u6210\u4e24\u7c7b\u4efb\u52a1\u3002\u4ee5 CLIP-L \u89c6\u89c9\u7f16\u7801\u5668\u8f93\u51fa 1024 \u7ef4\u7279\u5f81\u3001\u5355\u5c42\u7ebf\u6027\u6295\u5f71\u6620\u5c04\u5230 4096 \u7ef4\u8bcd\u5d4c\u5165\u7684\u914d\u7f6e\u4e3a\u4f8b\uff0c\u6295\u5f71\u4ec5\u589e\u52a0\u7ea6 4.2M \u53c2\u6570\uff0c\u8bad\u7ec3\u663e\u5b58\u76f8\u6bd4\u7eaf\u6587\u672c\u591a 30% \u5230 50%\u3002","og_url":"https:\/\/www.wsisp.com\/helps\/92884.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-08-11T05:56:48+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260811055646-6a7ab99ee25b3.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"7 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/92884.html","url":"https:\/\/www.wsisp.com\/helps\/92884.html","name":"\u591a\u6a21\u6001\u8bad\u7ec3\uff1a\u4ece\u50cf\u7d20\u5230\u8bed\u4e49\uff0c\u4e00\u4e2a\u6a21\u578b\u7edf\u6cbb\u6240\u6709\u6a21\u6001 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-08-11T05:56:48+00:00","dateModified":"2026-08-11T05:56:48+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/92884.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/92884.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/92884.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u591a\u6a21\u6001\u8bad\u7ec3\uff1a\u4ece\u50cf\u7d20\u5230\u8bed\u4e49\uff0c\u4e00\u4e2a\u6a21\u578b\u7edf\u6cbb\u6240\u6709\u6a21\u6001"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/92884","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=92884"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/92884\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/92883"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=92884"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=92884"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=92884"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=92884"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}