
{"id":11461,"date":"2026-08-26T16:41:03","date_gmt":"2026-08-26T08:41:03","guid":{"rendered":"https:\/\/infernews.com\/blog\/reward-modeling-from-human-preference-feedback-for-joint-video-audio-generation\/"},"modified":"2026-08-26T16:41:03","modified_gmt":"2026-08-26T08:41:03","slug":"reward-modeling-from-human-preference-feedback-for-joint-video-audio-generation","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/reward-modeling-from-human-preference-feedback-for-joint-video-audio-generation\/","title":{"rendered":"VA-Judger \u751f\u6210\u66f4\u8cbc\u8fd1\u4eba\u985e\u504f\u597d\u7684\u5f71\u7247\u8207\u8072\u97f3"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/pasted-495de6a3621b.jpg\" alt=\"VA-Judger training pipeline\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u540c\u4e00\u500b\u63d0\u793a\u8a5e\u7522\u751f\u5169\u6bb5\u5f71\u97f3\u5167\u5bb9\u6642\uff0cVA-Judger\u6703\u6bd4\u8f03\u54ea\u4e00\u6bb5\u66f4\u7b26\u5408\u4eba\u985e\u504f\u597d\uff0c\u4e26\u62c6\u89e3\u63d0\u793a\u8a5e\u5c0d\u9f4a\u3001\u5f71\u97f3\u4e00\u81f4\u6027\u3001\u97f3\u8cea\u3001\u756b\u8cea\u53ca\u5167\u5bb9\u5b8c\u6574\u5ea6\u3002\u5b83\u5c6c\u65bc\u806f\u5408\u5f71\u7247\u8207\u97f3\u8a0a\u751f\u6210\u7684 Reward Model\uff0c\u8655\u7406\u50b3\u7d71\u6307\u6a19\u96e3\u4ee5\u6355\u6349\u6574\u9ad4\u9023\u8cab\u6027\u7684\u554f\u984c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6a21\u578b\u4ee5 Qwen3-Omni \u70ba\u57fa\u790e\uff0c\u5148\u5f9e\u54c1\u8cea\u5dee\u8ddd\u660e\u986f\u7684\u914d\u5c0d\u5b78\u7fd2\u6bd4\u8f03\u6e96\u5247\uff0c\u518d\u900f\u904e\u62d2\u7d55\u63a1\u6a23\u8655\u7406\u63a5\u8fd1\u7684\u7d50\u679c\uff0c\u6700\u5f8c\u4ee5 dimension-wise Group Relative Policy Optimization\uff08GRPO\uff09\u628a\u4eba\u985e\u56de\u994b\u5206\u914d\u5230\u4e0d\u540c\u54c1\u8cea\u7dad\u5ea6\u3002\u9019\u6bd4\u628a\u97f3\u8cea\u3001\u756b\u8cea\u53ca\u540c\u6b65\u6307\u6a19\u7c21\u55ae\u76f8\u52a0\u66f4\u8cbc\u8fd1\u89c0\u770b\u8005\u7684\u6574\u9ad4\u5224\u65b7\uff0c\u4e5f\u6e1b\u5c11\u751f\u6210\u6a21\u578b\u947d\u6307\u6a19\u6f0f\u6d1e\u7684\u6a5f\u6703\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">VA-Judger-Bench\u5305\u542b\u540c\u9818\u57df\u53ca\u8de8\u9818\u57df\u6a21\u578b\u6bd4\u8f03\uff0c\u7528\u4f86\u6e2c\u8a66 Reward Model \u662f\u5426\u80fd\u5c0d\u9f4a\u4eba\u985e\u9078\u64c7\uff1b\u7814\u7a76\u7d50\u679c\u6307\u5411\u5b83\u512a\u65bc\u591a\u9805\u6307\u6a19\u57fa\u7dda\u3002VAPref-10K\u5247\u5305\u542b9K\u63d0\u793a\u8a5e\u53ca10.3K\u7d44\u7d30\u7dfb\u7684\u5f71\u97f3\u914d\u5c0d\u6bd4\u8f03\uff0c\u4f46\u8cc7\u6599\u96c6\u53ca\u90e8\u5206\u8a13\u7df4\u7a0b\u5f0f\u4ecd\u672a\u767c\u5e03\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9805\u76ee\u63d0\u4f9b Reward Model \u63a8\u7406\u3001Video Model \u63a8\u7406\u3001\u6a21\u578b SFT\uff0c\u4ee5\u53ca\u4ee5 VA-Judger \u5206\u6578\u512a\u5316 LTX-2 \u7684\u6d41\u7a0b\uff1bREADME\u4ea6\u5217\u51fa\u5408\u4f75\u5df2\u767c\u5e03 RL LoRA \u7684 LTX-2 checkpoint\u3002\u63d0\u4f9b\u7684\u8cc7\u6599\u6c92\u6709\u5217\u51fa\u5b8c\u6574\u5b89\u88dd\u6b65\u9a5f\uff0c\u4f7f\u7528\u8005\u9700\u8981\u6309\u9805\u76ee\u9801\u9762\u3001checkpoint\u53ca\u7a0b\u5f0f\u78bc\u72c0\u614b\u81ea\u884c\u78ba\u8a8d\u74b0\u5883\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u9069\u5408\u5834\u666f<\/strong>\uff1a\u7814\u7a76\u5f71\u97f3\u751f\u6210\u3001\u504f\u597d\u5c0d\u9f4a\u53ca RL post-training \u7684\u5718\u968a<\/li>\n<li><strong>\u53ef\u8a55\u4f30\u5167\u5bb9<\/strong>\uff1a\u63d0\u793a\u8a5e\u5c0d\u9f4a\u3001\u5f71\u97f3\u4e00\u81f4\u6027\u3001\u97f3\u8cea\u3001\u756b\u8cea\u53ca\u5167\u5bb9\u5b8c\u6574\u5ea6<\/li>\n<li><strong>\u4e3b\u8981\u53d6\u6368<\/strong>\uff1a\u6bd4\u8f03\u4eba\u985e\u504f\u597d\u7684\u65b9\u6cd5\u8f03\u5168\u9762\uff0c\u4f46\u9700\u8981\u914d\u5c0d\u8cc7\u6599\u53ca\u984d\u5916\u8a13\u7df4\u6d41\u7a0b<\/li>\n<li><strong>\u76ee\u524d\u9650\u5236<\/strong>\uff1adimension-wise GRPO \u7a0b\u5f0f\u78bc\u53ca VAPref-10K \u4ecd\u5217\u70ba\u5f85\u767c\u5e03\u9805\u76ee<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/sharelab-sii.github.io\/VA-Judger\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/ShareLab-SII\/VA-Judger\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a> \u00b7 <a href=\"https:\/\/huggingface.co\/ShareLab-SII\/VA-Judger\/tree\/main\" rel=\"noopener noreferrer\"><strong>\u6a21\u578b<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>VA-Judger \u4ee5\u4eba\u985e\u504f\u597d\u6bd4\u8f03\u5f71\u7247\u548c\u8072\u97f3\u751f\u6210\u7d50\u679c\uff0c\u6539\u5584\u55ae\u9760\u7368\u7acb\u6307\u6a19\u9020\u6210\u7684\u5931\u771f\u8207\u734e\u52f5\u932f\u914d\u3002<\/p>\n","protected":false},"author":8,"featured_media":11460,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,185,31,76,127,149,141,194,199],"tags":[],"class_list":["post-11461","post","type-post","status-publish","format-standard","hentry","category-133","category-qwen","category-video","category-76","category-127","category-149","category-141","category-ltx","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11461","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=11461"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11461\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/11460"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=11461"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=11461"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=11461"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}