{"id":11974,"date":"2026-09-07T22:00:27","date_gmt":"2026-09-07T14:00:27","guid":{"rendered":"https:\/\/infernews.com\/blog\/a-unified-omni-evaluation-framework\/"},"modified":"2026-09-07T22:00:27","modified_gmt":"2026-09-07T14:00:27","slug":"a-unified-omni-evaluation-framework","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/a-unified-omni-evaluation-framework\/","title":{"rendered":"OmniEvalKit\uff1a\u5514\u4f7f\u91cd\u65b0\u8a13\u7df4 VLM \u90fd\u53ef\u4ee5\u807d\u8072\u7b54\u554f\u984c"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/09\/TFO_logo.jpg\" alt=\"Training-Free Omni\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u60f3\u70ba\u4e00\u500b\u8996\u89ba\u8a9e\u8a00\u6a21\u578b\u52a0\u5165\u8a9e\u97f3\u7406\u89e3\uff0c\u4f46\u53c8\u5514\u60f3\u91cd\u65b0\u8a13\u7df4\uff1fMBZUAI Oryx \u5718\u968a\u958b\u6e90\u5605 OmniEvalKit\uff08Training-Free Omni\uff09\u5c31\u6b63\u6b63\u91dd\u5c0d\u5462\u500b\u75db\u9ede\u3002\u5b83\u628a\u8a9e\u97f3\u5148\u7d93 Whisper \u62bd\u53d6\u6210\u5e36\u6642\u9593\u6233\u3001\u8a9e\u8a00\u540c\u4fe1\u5fc3\u5206\u6578\u5605\u7d50\u69cb\u5316\u6587\u5b57\uff0c\u518d\u9023\u540c\u5716\u7247\u6216\u5f71\u7247\u5e40\u4e00\u9f4a\u9935\u4ffe\u51cd\u7d50\u5605 VLM\uff0c\u6240\u6709\u63a8\u7406\u90fd\u6cbf\u7528\u539f\u672c\u5605 prompt \u63a5\u53e3\uff0c\u9aa8\u5e79\u6b0a\u91cd\u5168\u7a0b\u4e0d\u52d5\u3002\u63db\u53e5\u8a71\u8b1b\uff0c\u4efb\u4f55\u65b0\u5605\u8996\u89ba\u9aa8\u5e79\u90fd\u53ef\u4ee5\u5373\u63d2\u5373\u7528\uff0c\u6bcb\u9808\u518d\u505a\u8a9e\u97f3\u2014\u8996\u89ba\u5c0d\u9f4a\u5fae\u8abf\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b83\u540c\u6642\u4fc2\u4e00\u500b\u7d71\u4e00\u5605\u591a\u6a21\u614b\u8a55\u6e2c\u6846\u67b6\uff0c\u652f\u63f4\u6587\u5b57\u3001\u5716\u7247\u3001\u5f71\u7247\u3001\u97f3\u983b\u540c\u97f3\u8996\u983b\u4efb\u52d9\uff0c\u4e26\u9810\u8f09 118 \u500b\u8cc7\u6599\u96c6\u9069\u914d\u5668\uff0c\u6db5\u84cb Qwen\u3001Gemma\u3001MiniCPM\u3001VILA\u3001OmniVinci \u7b49\u6a21\u578b\uff0c\u65b9\u4fbf\u505a\u516c\u5e73\u5c0d\u9f4a\u6e2c\u8a66\u3002\u5c0d\u7814\u7a76\u4eba\u54e1\u540c\u90e8\u7f72\u5718\u968a\u800c\u8a00\uff0c\u6700\u76f4\u63a5\u5605\u597d\u8655\u4fc2\u53ef\u4ee5\u4e00\u6b21\u904e\u8dd1 56 \u500b benchmark\u300121 \u7a2e\u8a9e\u8a00\uff0c\u76f4\u63a5\u6bd4\u8f03\u51cd\u7d50\u9aa8\u5e79\u540c\u539f\u751f omni \u6a21\u578b\u4e4b\u9593\u5605\u5dee\u8ddd\uff0c\u7747\u4e0b\u8a9e\u97f3\u80fd\u529b\u7a76\u7adf\u4fc2\u65b0\u52a0\u51fa\u569f\u5b9a\u4fc2\u7531\u820a\u80fd\u529b\u4ea4\u63db\u5f97\u569f\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5982\u679c\u4f60\u95dc\u5fc3 VLM \u52a0\u639b\u8a9e\u97f3\u5f8c\u6703\u5514\u6703\u300c\u5931\u61b6\u300d\uff0c\u5462\u5957\u6846\u67b6\u6b63\u6b63\u63d0\u4f9b matched comparison\uff0c\u53ef\u4ee5\u91cf\u5316\u8a55\u4f30\u5716\u50cf\u7406\u89e3\u3001\u8996\u89ba\u5b9a\u4f4d\u3001\u7de8\u78bc\u3001\u6578\u5b78\u7b49\u539f\u6709\u5f37\u9805\u6709\u5187\u88ab\u524a\u5f31\u3002\u984d\u5916\u652f\u63f4\u5605 CosyVoice3 \u6587\u5b57\u8f49\u8a9e\u97f3\u8f38\u51fa\uff0c\u4ea6\u4ee4\u6587\u672c\u7b54\u6848\u53ef\u4ee5\u76f4\u63a5\u8b8a\u6210\u8a9e\u97f3\u56de\u8986\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8981\u672c\u5730\u8dd1\u5f97\u8d77\u569f\uff0c\u9700\u8981 Python 3.10+\u3001ffmpeg\uff0c\u518d\u91dd\u5c0d CPU\u3001CUDA \u6216 ROCm \u5b89\u88dd\u5c0d\u61c9\u5605 PyTorch\u3002\u4e4b\u5f8c\u900f\u904e eval.sh \u914d\u74b0\u5883\u8b8a\u6578\u6307\u5b9a\u6a21\u578b\u540c\u8cc7\u6599\u96c6\u5373\u53ef\u958b\u8dd1\uff0c\u52a0 MAX_SAMPLES=3 \u53ef\u4ee5\u505a\u7159\u9727\u6e2c\u8a66\uff0c\u4e2d\u65b7\u5f8c\u8a2d RESUME=True \u53ef\u4ee5\u63a5\u8fd4\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4ee5\u4e0b\u4fc2\u5e7e\u500b\u503c\u5f97\u7559\u610f\u5605\u91cd\u9ede\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u51cd\u7d50\u9aa8\u5e79\u3001\u96f6\u5fae\u8abf<\/strong>\uff1a\u6240\u6709 VLM \u6b0a\u91cd\u5b8c\u5168\u4e0d\u8b8a\uff0c\u8a9e\u97f3\u7406\u89e3\u900f\u904e Whisper \u62bd\u53d6\u6587\u5b57\u8b49\u64da\u518d\u52a0 prompt \u878d\u5408\u9054\u6210\u3002<\/li>\n<li><strong>\u5373\u63d2\u5373\u7528\u5605 omni \u80fd\u529b<\/strong>\uff1a\u652f\u63f4 Qwen2.5-Omni\u3001Gemma\u3001MiniCPM\u3001VILA\u3001OmniVinci \u7b49\u591a\u500b\u6a21\u578b\u9069\u914d\u5668\uff0c\u65b9\u4fbf\u6a2a\u5411\u6bd4\u8f03\u3002<\/li>\n<li><strong>\u8986\u84cb\u5ee3\u5605\u8a55\u6e2c\u77e9\u9663<\/strong>\uff1a\u5167\u7f6e 118 \u500b\u8cc7\u6599\u96c6\u9069\u914d\u5668\uff0c\u6db5\u84cb 56 \u500b benchmark \u8207 21 \u7a2e\u8a9e\u8a00\u3002<\/li>\n<li><strong>\u539f\u751f omni \u540c\u51cd\u7d50\u9aa8\u5e79\u5605 matched \u5c0d\u7167<\/strong>\uff1a\u53ef\u4ee5\u6e05\u6670\u5206\u8fa8\u65b0\u589e\u80fd\u529b\u540c\u4fdd\u7559\u80fd\u529b\uff0c\u907f\u514d\u91cd\u8a13\u5e36\u569f\u5605 capability drift\u3002<\/li>\n<li><strong>\u53ef\u9078\u8a9e\u97f3\u56de\u8986<\/strong>\uff1a\u900f\u904e CosyVoice3 \u628a\u6587\u5b57\u7b54\u6848\u5408\u6210\u8a9e\u97f3\u8f38\u51fa\uff0c\u9069\u5408\u5c0d\u8a71\u5f0f\u5834\u666f\u3002<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/mbzuai-oryx.github.io\/OmniEvalKit\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/mbzuai-oryx\/OmniEvalKit\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>MBZUAI Oryx \u5718\u968a\u628a OmniEvalKit \u958b\u6e90\u51fa\u569f\uff0c\u4e3b\u6253\u5514\u4f7f\u6539\u52d5 VLM \u4efb\u4f55\u53c3\u6578\uff0c\u5c31\u70ba\u4f62\u52a0\u639b\u8a9e\u97f3\u7406\u89e3\u80fd\u529b\u3002\u5c0d\u65bc\u60f3\u8a55\u4f30\u6216\u90e8\u7f72\u591a\u6a21\u614b\u6a21\u578b\u5605\u5718\u968a\uff0c\u53ef\u4ee5\u76f4\u63a5\u62ce\u73fe\u6210\u9aa8\u5e79\u5373\u8a66\u3002<\/p>\n","protected":false},"author":8,"featured_media":11973,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,185,140,179,37,31,30,162,152,119,76,127,149,128,197,199],"tags":[],"class_list":["post-11974","post","type-post","status-publish","format-standard","hentry","category-133","category-qwen","category-gemini","category-nvidia","category-37","category-video","category-image","category-ai-productions","category-python","category-119","category-76","category-127","category-149","category-128","category-framework","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11974","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=11974"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11974\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/11973"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=11974"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=11974"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=11974"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}