{"id":10870,"date":"2026-08-10T21:21:49","date_gmt":"2026-08-10T13:21:49","guid":{"rendered":"https:\/\/infernews.com\/blog\/streamarena\/"},"modified":"2026-08-10T21:26:58","modified_gmt":"2026-08-10T13:26:58","slug":"streamarena","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/streamarena\/","title":{"rendered":"StreamArena \u591a\u6a21\u614b\u9577\u7247\u4ee3\u7406\u7684\u8a18\u61b6\u8207\u4e92\u52d5\u5206\u6790"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/pasted-8defa277fb94.jpg\" alt=\"StreamArena overview\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u7531 HKUST(\u9999\u6e2f\u79d1\u6280\u5927\u5b78)\u3001The University of Hong Kong(\u9999\u6e2f\u5927\u5b78) \u8207 The Chinese University of Hong Kong(\u9999\u6e2f\u4e2d\u6587\u5927\u5b78) \u53ca\u5c0f\u7d05\u66f8\u7d44\u6210\u7684\u958b\u767c\u5718\u968a\uff0c\u5c07\u7126\u9ede\u653e\u5230\u4e00\u500b\u5f88\u591a\u591a\u6a21\u614b\u4ee3\u7406\u90fd\u672a\u771f\u6b63\u8655\u7406\u597d\u7684\u96e3\u984c\uff1a\u5f71\u7247\u5514\u518d\u4fc2\u5e7e\u79d2\u526a\u8f2f\uff0c\u800c\u4fc2\u63a5\u8fd1\u4e00\u500b\u534a\u9418\u982d\u3001\u4ef2\u8981\u6301\u7e8c\u4e92\u52d5\u3002StreamArena \u5c6c\u65bc<strong>\u8cc7\u6599\u96c6\u8207\u8a55\u6e2c\u5de5\u5177\u5305<\/strong>\uff0c\u8655\u7406\u7684\u662f continuous streaming video understanding\uff0c\u76ee\u6a19\u4fc2\u7528\u7d71\u4e00\u65b9\u6cd5\u6aa2\u9a57\u4ee3\u7406\u5728\u9577\u6642\u9593\u5f71\u97f3\u6d41\u4e4b\u4e2d\uff0c\u80fd\u5426\u5373\u6642\u7406\u89e3\u3001\u9694\u4e00\u6bb5\u6642\u9593\u5f8c\u4ecd\u7136\u8a18\u5f97\u5167\u5bb9\uff0c\u4e26\u5728\u5408\u9069\u6642\u6a5f\u7528\u5de5\u5177\u6216\u4e3b\u52d5\u56de\u61c9\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5462\u500b\u9805\u76ee\u6700\u503c\u5f97\u6ce8\u610f\u7684\u5730\u65b9\uff0c\u5728\u65bc\u5b83\u5514\u7528\u77ed\u7247\u52a0\u9078\u64c7\u984c\u53bb\u63a9\u84cb\u6a21\u578b\u5f31\u9ede\uff0c\u800c\u4fc2\u7528 243 \u6bb5 full-length videos\u3001\u5e73\u5747 88.8 \u5206\u9418\u3001\u5408\u5171 3,646 \u500b open-ended tasks\uff0c\u5206\u6210 Real-Time Perception (RTP)\u3001Historical Retrospection (HR)\u3001External Tool Use (Tool) \u540c Proactive Interaction (Pro)\u3002HR \u8207 Pro \u4ef2\u6309\u6642\u9593\u8de8\u5ea6\u5206\u5c64\uff0cHR \u6700\u9060\u62c9\u5230 30 \u5206\u9418\uff0c\u76f4\u63a5\u628a\u9577\u671f\u8a18\u61b6\u8207\u5ef6\u9072\u63a7\u5236\u7684\u53d6\u6368\u6524\u958b\u4f86\u6e2c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">StreamArena Evaluation Toolkit\uff1a<code>streamarena\/<\/code> \u653e\u8cc7\u6599\u8f09\u5165\u3001\u5a92\u9ad4\u8655\u7406\u3001tool-call protocol\u3001OpenAIBackend\u3001GeminiBackend \u540c LLM-as-Judge scorer\uff1b<code>method\/<\/code> \u5247\u6574\u7406\u591a\u7a2e\u88ab\u6e2c\u65b9\u6cd5\uff0c\u5305\u62ec offline \u7684 Qwen\u3001MiMo\u3001Kimi\u3001Qwen-Omni\u3001Gemini\uff0c\u4ee5\u53ca MiniCPM-o-4.5\u3001VST\u3001StreamForest\u3001ThinkStream\u3002\u6240\u6709\u65b9\u6cd5\u90fd\u5beb\u5165\u540c\u4e00\u5957 records JSONL schema\uff0c\u6240\u4ee5 <code>judge\/<\/code> \u53ef\u4ee5\u7528\u540c\u4e00\u628a\u5c3a\u8a55\u5206\u3002\u5b89\u88dd\u8207\u57f7\u884c\u7d30\u7bc0\u5728\u76ee\u524d\u8cc7\u8a0a\u88e1\u672a\u5b8c\u5168\u5c55\u958b\uff0c\u4f46\u53ef\u78ba\u8a8d\u5b83\u4e0d\u662f\u55ae\u4e00\u6a21\u578b\u5009\u5eab\uff0c\u800c\u662f\u7528\u4f86\u91cd\u8dd1\u3001\u5c0d\u9f4a\u8207\u6bd4\u8f03\u4e0d\u540c\u65b9\u6cd5\u7684\u8a55\u6e2c\u6846\u67b6\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8986\u84cb\u56db\u985e\u80fd\u529b\uff1aRTP\u3001HR\u3001Tool\u3001Pro\uff0c\u5514\u53ea\u554f\u7b54\uff0c\u4ef2\u6e2c\u6301\u7e8c\u76e3\u770b\u8207\u4e3b\u52d5\u4e92\u52d5<\/li>\n<li>243 \u6bb5\u9577\u5f71\u7247\u3001\u5e73\u5747 88.8 \u5206\u9418\uff0c\u8cc7\u6599\u898f\u6a21\u660e\u986f\u504f\u5411\u9577\u6642\u5e8f\u7406\u89e3<\/li>\n<li>\u540c\u4e00\u4efd JSONL \u8f38\u51fa\u683c\u5f0f\u914d\u5408\u901a\u7528 judge\uff0c\u65b9\u4fbf\u6a6b\u5411\u6bd4\u8f03\u4e0d\u540c\u65b9\u6cd5<\/li>\n<li>StreamMind \u5728\u540c\u4e00 Qwen3.5-397B-A17B backbone \u4e0a\uff0c\u628a pooled query-to-answer latency \u964d\u4f4e 66.2%<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e00\u500b\u91cd\u8981\u9650\u5236\uff1aStreamMind \u5728\u9019\u6b21\u91cb\u51fa\u88e1\u53ea\u6709 evaluation protocol docs\uff0c\u672a\u9644\u5b8c\u6574\u53ef\u76f4\u63a5\u91cd\u73fe\u7684\u5be6\u4f5c\uff1bAURA \u4ea6\u53ea\u63d0\u4f9b client\uff0cserver \u4ecd\u8981\u4f9d\u8cf4\u5b98\u65b9 vLLM\u3002\u63db\u53e5\u8a71\u8aaa\uff0cStreamArena \u73fe\u968e\u6bb5\u66f4\u50cf\u7814\u7a76\u5718\u968a\u3001\u4ee3\u7406\u7cfb\u7d71\u958b\u767c\u8005\u540c\u591a\u6a21\u614b\u8a55\u6e2c\u5de5\u4f5c\u6d41\u6703\u7528\u5230\u7684\u57fa\u790e\u8a2d\u65bd\u3002\u60f3\u6bd4\u8f03\u4e0d\u540c streaming method\u3001\u6aa2\u67e5\u9577\u5f71\u7247\u4ee3\u7406\u7a76\u7adf\u5361\u5728\u8a18\u61b6\u3001\u53cd\u61c9\uff0c\u9084\u662f\u5de5\u5177\u8abf\u7528\uff0c\u5462\u500b\u9805\u76ee\u6bd4\u4e00\u822c\u77ed\u7247 benchmark \u66f4\u63a5\u8fd1\u90e8\u7f72\u524d\u8981\u9762\u5c0d\u7684\u73fe\u5be6\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/hkuzxc.github.io\/StreamArena_web\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/JIA-Lab-research\/StreamArena\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u4f86\u81ea\u9999\u6e2f\u79d1\u5927\u3001\u6e2f\u5927\u8207\u4e2d\u5927\u53ca\u5c0f\u7d05\u66f8\u7684\u5718\u968a\uff0c\u628a\u9577\u9054\u8fd1 90 \u5206\u9418\u7684\u5f71\u7247\u7406\u89e3\u62c6\u6210\u53ef\u6301\u7e8c\u89c0\u5bdf\u3001\u56de\u60f3\u3001\u627e\u5de5\u5177\u8207\u4e3b\u52d5\u56de\u61c9\u56db\u7a2e\u80fd\u529b\u3002<\/p>\n","protected":false},"author":8,"featured_media":10869,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,185,163,169,164,165,140,116,31,119,207],"tags":[],"class_list":["post-10870","post","type-post","status-publish","format-standard","hentry","category-133","category-qwen","category-163","category-169","category-164","category-165","category-gemini","category-agentic","category-video","category-119","category-kimi"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10870","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=10870"}],"version-history":[{"count":2,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10870\/revisions"}],"predecessor-version":[{"id":10875,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10870\/revisions\/10875"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/10869"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=10870"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=10870"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=10870"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}