{"id":13463,"date":"2026-10-16T18:31:07","date_gmt":"2026-10-16T10:31:07","guid":{"rendered":"https:\/\/infernews.com\/blog\/edgewise-bringing-long-horizon-agents-to-the-edge-via-selective-cloud-delegation\/"},"modified":"2026-10-16T18:31:07","modified_gmt":"2026-10-16T10:31:07","slug":"edgewise-bringing-long-horizon-agents-to-the-edge-via-selective-cloud-delegation","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/edgewise-bringing-long-horizon-agents-to-the-edge-via-selective-cloud-delegation\/","title":{"rendered":"EdgeWise \u8b93\u9577\u6d41\u7a0b AI \u4ee3\u7406\u5728\u672c\u5730\u904b\u4f5c\u66f4\u7701\u6210\u672c"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/10\/EdgeWise-AI.jpg\" alt=\"Og image\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u9577\u6d41\u7a0b AI \u4ee3\u7406\u4e0d\u5fc5\u6bcf\u4e00\u6b65\u90fd\u4f9d\u8cf4\u6602\u8cb4\u7684\u96f2\u7aef\u6a21\u578b\u3002NVIDIA \u7684 EdgeWise \u662f\u4e00\u5957 edge-cloud collaboration framework\uff0c\u8b93\u672c\u5730 SLM\uff08Small Language Model\uff09\u5148\u8ca0\u8cac\u89e3\u984c\uff0c\u9047\u5230\u9700\u8981\u66f4\u5f37\u63a8\u7406\u80fd\u529b\u7684\u95dc\u9375\u6b65\u9a5f\uff0c\u624d\u9078\u64c7\u6027\u5730\u4ea4\u7531 cloud LLM \u8655\u7406\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">EdgeWise \u7528 step-level RL\uff08Reinforcement Learning\uff09\u540c\u6642\u8a13\u7df4\u4efb\u52d9\u57f7\u884c\u8207\u81ea\u52d5\u5206\u5de5\uff0c\u4ee4\u4ee3\u7406\u81ea\u884c\u5224\u65b7\u4f55\u6642\u9700\u8981\u5916\u63f4\uff0c\u800c\u4e14\u4e0d\u5fc5\u5728\u63a8\u7406\u968e\u6bb5\u984d\u5916\u52a0\u5165\u53e6\u4e00\u500b\u6c7a\u7b56\u6a21\u578b\u3002\u7814\u7a76\u4ea6\u4ee5 external judge-based delegation strategy \u5206\u6790\u59d4\u6d3e\u6642\u6a5f\u3001\u5206\u4eab\u591a\u5c11\u4e0a\u4e0b\u6587\uff0c\u4ee5\u53ca\u96f2\u7aef\u6a21\u578b\u61c9\u8a72\u591a\u4e45\u4ecb\u5165\u4e00\u6b21\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u7814\u7a76\u767c\u73fe\uff0c\u96f2\u7aef\u667a\u6167\u9700\u6c42\u901a\u5e38\u96c6\u4e2d\u5728\u5c11\u6578\u95dc\u9375\u6b65\u9a5f\uff0c\u547c\u53eb\u6b21\u6578\u4e0d\u4e00\u5b9a\u8d8a\u591a\u8d8a\u597d\uff1b\u96a8\u6a5f\u628a\u7d04 40% \u6b65\u9a5f\u4ea4\u7d66\u96f2\u7aef\uff0c\u6709\u6642\u5df2\u80fd\u63a5\u8fd1\u5168\u96f2\u7aef\u57f7\u884c\u7684\u8868\u73fe\u3002EdgeWise \u5728 SWE-Rebench \u6e1b\u5c11\u6700\u591a 62% inference cost\uff0c\u5728 Terminal-Bench 2.1 \u6e1b\u5c11\u6700\u591a 34%\uff0c\u4e26\u5728\u672c\u5730\u5de5\u4f5c\u7ad9\u5b8c\u6210 edge-cloud \u6e2c\u8a66\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u2022 \u672c\u5730 SLM \u8ca0\u8cac\u4e3b\u8981\u4efb\u52d9\u6d41\u7a0b<br \/>\n\u2022 \u53ea\u5728\u95dc\u9375\u6b65\u9a5f\u59d4\u6d3e cloud LLM<br \/>\n\u2022 \u4ee5 step-level RL \u5b78\u7fd2\u59d4\u6d3e\u6c7a\u7b56<br \/>\n\u2022 \u652f\u63f4\u4e0d\u540c agent harnesses<br \/>\n\u2022 \u6210\u672c\u8f03\u5168\u96f2\u7aef\u57f7\u884c\u6700\u591a\u4f4e 62%<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9069\u5408\u9700\u8981\u8655\u7406\u9577\u6d41\u7a0b\u7de8\u7a0b\u3001\u7d42\u7aef\u6a5f\u64cd\u4f5c\u6216\u5176\u4ed6\u591a\u6b65\u9a5f\u4efb\u52d9\u7684\u5de5\u4f5c\u6d41\uff0c\u5c24\u5176\u9069\u5408\u5e0c\u671b\u964d\u4f4e inference cost\u3001\u53c8\u4e0d\u60f3\u5b8c\u5168\u72a7\u7272\u6a21\u578b\u80fd\u529b\u7684\u672c\u5730\u8a2d\u5099\u3002\u96f2\u7aef\u9023\u7dda\u3001\u59d4\u6d3e\u5224\u65b7\u548c\u4e0a\u4e0b\u6587\u5206\u4eab\u65b9\u5f0f\u4ecd\u6703\u5f71\u97ff\u7d50\u679c\uff0c\u56e0\u6b64\u5b83\u4e26\u975e\u5b8c\u5168\u812b\u96e2\u96f2\u7aef\u7684\u65b9\u6848\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/research.nvidia.com\/labs\/lpr\/edge-wise\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>EdgeWise \u8b93\u672c\u5730\u6a21\u578b\u8655\u7406\u5927\u90e8\u5206\u9577\u6d41\u7a0b\u4efb\u52d9\uff0c\u53ea\u5728\u95dc\u9375\u6b65\u9a5f\u547c\u53eb\u96f2\u7aef LLM\uff0c\u517c\u9867\u8868\u73fe\u8207\u6210\u672c\u3002<\/p>\n","protected":false},"author":8,"featured_media":13462,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","wpai_meta_description":"","footnotes":""},"categories":[116,76,127,179,210,197,146,214],"tags":[],"class_list":["post-13463","post","type-post","status-publish","format-standard","hentry","category-agentic","category-76","category-127","category-nvidia","category-worldactionmodel","category-framework","category-146","category-reinforcement-learning"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/13463","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=13463"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/13463\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/13462"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=13463"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=13463"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=13463"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}