
{"id":8431,"date":"2026-05-21T19:18:42","date_gmt":"2026-05-21T11:18:42","guid":{"rendered":"https:\/\/infernews.com\/blog\/code-for-when-vision-speaks-for-sound\/"},"modified":"2026-05-21T19:18:42","modified_gmt":"2026-05-21T11:18:42","slug":"code-for-when-vision-speaks-for-sound","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/code-for-when-vision-speaks-for-sound\/","title":{"rendered":"wvs-code\uff1a\u7528\u5f71\u7247\u8207\u8072\u97f3\u9a57\u8b49\u6a21\u578b\u662f\u5426\u771f\u61c2\u5167\u5bb9"},"content":{"rendered":"<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/05\/pasted-de668f8fc50f.jpg\" alt=\"Repository image for rakanWen\/wvs-code\"><\/figure>\n<p>\u5c08\u6848\u300aWhen Vision Speaks for Sound\u300bt\u63d0\u4f9b\u5b98\u65b9\u7a0b\u5f0f\u78bc\uff0c\u6838\u5fc3\u76ee\u6a19\u4e0d\u662f\u505a\u4e00\u822c\u5f71\u97f3\u554f\u7b54\uff0c\u800c\u662f\u6aa2\u67e5\u652f\u63f4\u5f71\u7247\u7684\u591a\u6a21\u614b\u6a21\u578b\uff0c\u7a76\u7adf\u6709\u6c92\u6709\u771f\u6b63\u7406\u89e3\u8072\u97f3\uff0c\u9084\u662f\u53ea\u9760\u756b\u9762\u548c\u8a9e\u610f\u731c\u7b54\u6848\u3002\u5b83\u63d0\u4f9b\u6a21\u578b\u3001\u8a55\u4f30\u4ecb\u9762\u548c\u8a13\u7df4\u6d41\u7a0b\uff0c\u65b9\u4fbf\u7814\u7a76\u8005\u91cd\u73fe\u5be6\u9a57\u6216\u6539\u9020\u81ea\u5df1\u7684\u6e2c\u8a66\u65b9\u5f0f\u3002<\/p>\n<p>\u5118\u7ba1\u652f\u63f4\u8996\u983b\u7684\u591a\u6a21\u614b\u5927\u8a9e\u8a00\u6a21\u578b\uff08video-capable MLLMs\uff09\u9032\u6b65\u5f88\u5feb\uff0c\u4f46\u7814\u7a76\u767c\u73fe\u5b83\u5011\u5728\u8996\u983b\u4e2d\u8868\u73fe\u51fa\u7684\u300c\u97f3\u983b\u7406\u89e3\u300d\u80fd\u529b\u5f80\u5f80\u662f\u7531\u8996\u89ba\u9a45\u52d5\u7684\uff1a\u6a21\u578b\u5176\u5be6\u662f\u4f9d\u9760\u8996\u89ba\u7dda\u7d22\u4f86\u63a8\u65b7\u3001\u751a\u81f3\u5e7b\u60f3\u51fa\u8072\u97f3\u76f8\u95dc\u7684\u8cc7\u8a0a\uff0c\u800c\u4e0d\u662f\u771f\u6b63\u53bb\u6aa2\u67e5\u6216\u5206\u6790\u97f3\u983b\u4e32\u6d41\u672c\u8eab \u3002<\/p>\n<p>\u9019\u500b\u554f\u984c\u666e\u904d\u5b58\u5728\u65bc\uff1a<\/p>\n<p>\u6700\u5148\u9032\u7684\u958b\u6e90\u5168\u80fd\u6a21\u578b\uff08omni models\uff09<\/p>\n<p>\u4e3b\u8981\u9589\u6e90\u6a21\u578b\u4f9b\u61c9\u5546\uff08\u5982 Google \u548c OpenAI\uff09\u7684\u9802\u7d1a\u6a21\u578b<\/p>\n<p>\u63db\u53e5\u8a71\u8aaa\uff0c\u9019\u4e9b\u6a21\u578b\u770b\u8d77\u4f86\u80fd\u300c\u807d\u61c2\u300d\u8996\u983b\u4e2d\u7684\u8072\u97f3\uff0c\u4f46\u5be6\u969b\u4e0a\u5b83\u5011\u53ea\u662f\u300c\u770b\u300d\u756b\u9762\u4f86\u731c\u8072\u97f3\u662f\u4ec0\u9ebc\uff0c\u4e26\u6c92\u6709\u771f\u6b63\u8655\u7406\u97f3\u983b\u6578\u64da\uff0c\u56e0\u6b64\u5bb9\u6613\u7522\u751f\u932f\u8aa4\u6216\u5e7b\u89ba\uff08hallucinate\uff09\u3002<\/p>\n<p>\u5148\u6e96\u5099\u597d\u5f71\u7247\u548c\u97f3\u8a0a\u8cc7\u6599\uff0c\u518d\u628a\u8cc7\u6599\u767b\u8a18\u5230 LLaMA-Factory \u7684\u8cc7\u6599\u8a2d\u5b9a\u4e2d\uff0c\u4e4b\u5f8c\u5c31\u53ef\u4ee5\u7528\u5b83\u7684 SFT \u6216 DPO \u683c\u5f0f\u53bb\u8a13\u7df4\u3002\u5c08\u6848\u4e5f\u652f\u63f4\u628a\u6a23\u672c\u5beb\u6210 ShareGPT \u98a8\u683c\uff0c\u8b93\u6bcf\u689d\u8cc7\u6599\u540c\u6642\u5e36\u4e0a <code>&lt;video&gt;<\/code> \u548c <code>&lt;audio&gt;<\/code>\uff0c\u65b9\u4fbf\u6a21\u578b\u5b78\u7fd2\u5728\u591a\u6a21\u614b\u60c5\u5883\u4e0b\u4f5c\u7b54\u3002<\/p>\n<p>\u5b83\u6bd4\u8f03\u7279\u5225\u7684\u5730\u65b9\uff0c\u5728\u65bc\u63a1\u7528\u4ecb\u5165\u5f0f\u8a3a\u65b7\u6846\u67b6 Thud\uff0c\u5c08\u9580\u6e2c\u8a66\u6a21\u578b\u662f\u5426\u771f\u7684\u6709\u505a\u97f3\u8a0a\u9a57\u8b49\uff0c\u800c\u4e0d\u662f\u53ea\u8d70\u8996\u89ba\u6377\u5f91\u3002\u9019\u7a2e\u8a2d\u8a08\u5c0d\u7814\u7a76\u300c\u6a21\u578b\u5230\u5e95\u770b\u4e86\u751a\u9ebc\u3001\u807d\u4e86\u751a\u9ebc\u300d\u7279\u5225\u6709\u7528\uff0c\u4e5f\u6bd4\u55ae\u7d14\u6e96\u78ba\u7387\u66f4\u80fd\u63ed\u793a\u6a21\u578b\u884c\u70ba\u3002<\/p>\n<ul>\n<li>\u53ef\u7528\u4f86\u8a55\u6e2c\u5f71\u7247\u8a9e\u97f3\u3001\u97f3\u756b\u540c\u6b65\u3001\u6642\u9593\u5ef6\u9072\u7b49\u554f\u984c<\/li>\n<li>\u9069\u5408\u505a\u591a\u6a21\u614b\u6a21\u578b\u7814\u7a76\u3001\u9664\u932f\u548c\u57fa\u6e96\u6e2c\u8a66<\/li>\n<li>\u652f\u63f4 SFT \u8207 DPO \u8a13\u7df4\u6d41\u7a0b<\/li>\n<li>\u53ef\u63a5\u5165 LLaMA-Factory \u4e00\u8d77\u4f7f\u7528<\/li>\n<li>\u76f8\u95dc\u6a21\u578b\u8207\u6846\u67b6\u91cd\u9ede\u5305\u62ec Thud\u3001LLaMA-Factory \u4ee5\u53ca\u591a\u7a2e\u53ef\u8655\u7406\u5f71\u7247\u7684\u591a\u6a21\u614b\u6a21\u578b<\/li>\n<\/ul>\n<p>\u6574\u9ad4\u4f86\u8aaa\uff0c\u9019\u500b\u5c08\u6848\u66f4\u50cf\u662f\u4e00\u5957\u300c\u6aa2\u67e5\u5de5\u5177\u300d\uff0c\u800c\u4e0d\u662f\u9762\u5411\u4e00\u822c\u7528\u5bb6\u7684\u61c9\u7528\u7a0b\u5f0f\u3002\u5c0d\u505a AI \u7814\u7a76\u3001\u5f71\u97f3\u7406\u89e3\u8a55\u6e2c\uff0c\u6216\u8005\u60f3\u5206\u6790\u6a21\u578b\u6709\u6c92\u6709\u5077\u61f6\u9760\u756b\u9762\u731c\u7b54\u6848\u7684\u4eba\uff0c\u6703\u7279\u5225\u6709\u53c3\u8003\u50f9\u503c\u3002<\/p>\n<p><strong>GitHub\uff1a<\/strong> <a href=\"https:\/\/github.com\/rakanWen\/wvs-code\" rel=\"noopener noreferrer\">https:\/\/github.com\/rakanWen\/wvs-code<\/a><\/p>\n<p><strong>Paper\uff1a<\/strong> <a href=\"https:\/\/arxiv.org\/pdf\/2605.16403\" rel=\"noopener noreferrer\">https:\/\/arxiv.org\/pdf\/2605.16403<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Thud \u7528\u4ecb\u5165\u5f0f\u6e2c\u8a66\uff0c\u6aa2\u67e5\u591a\u6a21\u614b\u6a21\u578b\u662f\u9760\u771f\u5be6\u97f3\u8a0a\uff0c\u9084\u662f\u53ea\u770b\u756b\u9762\u7dda\u7d22\u3002\u9069\u5408\u505a\u7814\u7a76\u3001\u8a55\u6e2c\u8207\u8a13\u7df4\u8cc7\u6599\u6574\u7406\u3002<\/p>\n","protected":false},"author":8,"featured_media":8430,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[133,120,76,136,149,197],"tags":[],"class_list":["post-8431","post","type-post","status-publish","format-standard","hentry","category-133","category-120","category-76","category-136","category-149","category-framework"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8431","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=8431"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8431\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/8430"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=8431"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=8431"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=8431"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}