{"id":10757,"date":"2026-08-04T19:44:48","date_gmt":"2026-08-04T11:44:48","guid":{"rendered":"https:\/\/infernews.com\/blog\/deepvoyager-vl\/"},"modified":"2026-08-04T19:44:48","modified_gmt":"2026-08-04T11:44:48","slug":"deepvoyager-vl","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/deepvoyager-vl\/","title":{"rendered":"DeepVoyager-VL \u628a\u8996\u89ba\u7dda\u7d22\u653e\u56de\u641c\u5c0b\u6d41\u7a0b\u4e2d"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/pasted-48df27276663.jpg\" alt=\"Comparison of multimodal search data synthesis paradigms\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u7576\u4e00\u500b\u591a\u6a21\u614b agent \u8981\u9023\u7e8c\u67e5\u591a\u6b65\u8cc7\u8a0a\uff0c\u6700\u6613\u5931\u6e96\u5605\u4f4d\u7f6e\u5f80\u5f80\u5514\u4fc2\u7b54\u984c\uff0c\u800c\u4fc2\u4e2d\u9014\u6435\u932f\u7dda\u7d22\u3002DeepVoyager-VL\u8655\u7406\u5605\u6b63\u6b63\u4fc2\u5462\u500b\u554f\u984c\uff1a\u4f62\u5c6c\u65bc\u9577\u6d41\u7a0b\u591a\u6a21\u614b deep-search \u6846\u67b6\uff0c\u8b93\u65b0\u53d6\u5f97\u5605\u5716\u7247\u8b49\u64da\u76f4\u63a5\u5f71\u97ff\u4e0b\u4e00\u8f2a\u6aa2\u7d22\uff0c\u800c\u5514\u4fc2\u53ea\u55ba\u8f38\u5165\u958b\u982d\u6216\u7b54\u6848\u7d50\u5c3e\u5148\u7528\u5230\u8996\u89ba\u8cc7\u8a0a\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5462\u7a2e\u505a\u6cd5\u4ee4\u4f62\u540c\u4e00\u822c\u628a\u8996\u89ba\u8a0a\u606f\u5305\u88dd\u6210\u524d\u7f6e\u689d\u4ef6\u5605\u65b9\u6cd5\u6709\u660e\u986f\u5206\u5225\u3002DeepVoyager-VL\u80cc\u5f8c\u7528 EventVoyage-VL \u751f\u6210\u5e36\u6709\u4e2d\u9593\u8996\u89ba\u4f9d\u8cf4\u5605\u9577\u6d41\u7a0b\u554f\u984c\uff0c\u518d\u7528\u6574\u7406\u904e\u5605 trajectories \u505a Supervised Fine-Tuning\uff08SFT\uff09\uff0c\u800c\u5514\u52a0\u984d\u5916 reinforcement learning \u968e\u6bb5\uff1b\u53d6\u6368\u597d\u6e05\u695a\uff0c\u8a13\u7df4\u6d41\u7a0b\u8f03\u53ef\u91cd\u73fe\uff0c\u4f46\u6548\u679c\u4ecd\u7136\u8981\u9760\u8cc7\u6599\u5408\u6210\u8cea\u7d20\u540c\u8ecc\u8de1\u7be9\u9078\u6490\u4f4f\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5009\u5eab\u800c\u5bb6\u5df2\u7d93\u653e\u51fa paper\u3001project page\uff0c\u540c\u53ef\u91cd\u73fe\u5605 Megatron SFT training bundle\uff0c\u4ea6\u63d0\u4f9b DeepVoyager-VL-8B \u540c DeepVoyager-VL-30B-A3B \u6a21\u578b\uff0c\u4ee5\u53ca EventVoyage-VL dataset\u3002\u60f3\u7406\u89e3\u9ede\u6a23\u9a57\u8b49\uff0c\u6700\u76f4\u63a5\u4fc2\u6cbf\u4f4f\u73fe\u6210\u6a21\u578b\u3001\u8cc7\u6599\u96c6\u540c\u8a13\u7df4 bundle \u7747\u5b8c\u6574\u6d41\u7a0b\uff1bREADME \u4ea6\u63d0\u5230 <code>SWIFT RUNTIME=bundled<\/code> \u6703\u4f7f\u7528\u5132\u5b58\u5eab\u5167\u5605 source trees\uff0c\u5b9a\u4f4d\u4e0a\u6bd4\u8f03\u63a5\u8fd1\u7814\u7a76\u8207\u8a13\u7df4\u5fa9\u73fe\u9805\u76ee\uff0c\u800c\u5514\u4fc2\u5373\u958b\u5373\u7528\u5605\u6d88\u8cbb\u7d1a\u7522\u54c1\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u6838\u5fc3\u6539\u52d5<\/strong>\uff1a\u628a vision in the loop \u653e\u5165\u641c\u5c0b\u4e2d\u6bb5\uff0c\u5716\u7247\u53ef\u6309\u9700\u8981\u5148\u8f09\u5165\u6216\u88c1\u5207<\/li>\n<li><strong>\u8cc7\u6599\u4f86\u6e90<\/strong>\uff1aEventVoyage-VL \u5148\u505a structure-before-language synthesis\uff0c\u518d\u62bd\u51fa\u53ef\u76e3\u7763\u8ecc\u8de1<\/li>\n<li><strong>\u8a13\u7df4\u8def\u7dda<\/strong>\uff1a\u4ee5 supervised-only \u70ba\u4e3b\uff0c\u5187\u518d\u758a reinforcement learning \u968e\u6bb5<\/li>\n<li><strong>\u7d50\u679c<\/strong>\uff1a8B \u540c 30B-A3B \u5e73\u5747\u5206\u5225\u70ba 54.8 \u540c 58.6\uff0c\u4e26\u5728\u5341\u500b benchmark \u5165\u9762\u5206\u5225\u62ff\u4e0b\u516b\u500b\u540c\u4e5d\u500b\u6700\u4f73\u6210\u7e3e<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u5206\u6578\u672c\u8eab\u5df2\u7d93\u8b1b\u5230\u5b9a\u4f4d\uff1aDeepVoyager-VL\u4e0d\u53ea\u662f\u8ffd\u6c42\u66f4\u5927\u6a21\u578b\uff0c\u800c\u4fc2\u60f3\u6539\u5584\u300c\u898b\u5230\u65b0\u5716\u4e4b\u5f8c\uff0c\u4e0b\u4e00\u6b65\u61c9\u8a72\u6435\u4e5c\u300d\u5462\u500b\u6c7a\u7b56\u74b0\u7bc0\u3002\u8f03\u53d7\u76ca\u5605\u6703\u4fc2\u505a\u591a\u6b65\u6aa2\u7d22\u3001\u591a\u6a21\u614b\u554f\u7b54\u3001\u7814\u7a76\u578b agent pipeline \u5605\u5718\u968a\uff1b\u8981\u7559\u610f\u5605\u9650\u5236\u4ea6\u540c\u6a23\u76f4\u63a5\uff0c\u6210\u6548\u9ad8\u5ea6\u4f9d\u8cf4\u5408\u6210\u8cc7\u6599\u9ede\u6a23\u628a\u4e2d\u9014\u8996\u89ba\u4f9d\u8cf4\u7de8\u6392\u5f97\u5920\u771f\u5be6\uff0c\u96e2\u958b\u76f8\u95dc benchmark \u4e4b\u5f8c\uff0c\u6cdb\u5316\u6df1\u5ea6\u4ef2\u8981\u9760\u5f8c\u7e8c\u9a57\u8b49\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/halcyon-zhang.github.io\/DeepVoyager-VL\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/Halcyon-Zhang\/DeepVoyager-VL\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>DeepVoyager-VL\u4e0d\u53ea\u662f\u7747\u5716\u518d\u7b54\u984c\uff0c\u800c\u4fc2\u6703\u7528\u65b0\u53d6\u5f97\u5605\u8996\u89ba\u8b49\u64da\u6c7a\u5b9a\u4e0b\u4e00\u6b65\u641c\u5c0b\u3002\u5c0d\u9577\u6d41\u7a0b\u591a\u6a21\u614b\u6aa2\u7d22\u9805\u76ee\u800c\u8a00\uff0c\u5462\u500b\u6539\u52d5\u5e7e\u95dc\u9375\u3002<\/p>\n","protected":false},"author":8,"featured_media":10756,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,165,182,116,119,76,127,195,197,199],"tags":[],"class_list":["post-10757","post","type-post","status-publish","format-standard","hentry","category-133","category-165","category-182","category-agentic","category-119","category-76","category-127","category-195","category-framework","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10757","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=10757"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10757\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/10756"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=10757"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=10757"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=10757"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}