{"id":7868,"date":"2025-12-09T04:34:00","date_gmt":"2025-12-08T19:34:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=7868"},"modified":"2025-12-06T16:37:40","modified_gmt":"2025-12-06T07:37:40","slug":"arm-thinker-reinforcing-multimodal-generative-reward-models-with-agentic-tool-use-and-visual-reasoning","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=7868","title":{"rendered":"ARM-Thinker: Reinforcing Multimodal Generative Reward Models with Agentic Tool Use and Visual Reasoning\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>ARM-Thinker: Reinforcing Multimodal Generative Reward Models with Agentic Tool Use and Visual Reasoning\u00a0<\/strong>[103.8]<br>ARM-Thinker\u306f\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u30fb\u30de\u30eb\u30c1\u30e2\u30fc\u30c0\u30eb\u30fb\u30ea\u30ef\u30fc\u30c9\u30fb\u30e2\u30c7\u30eb\u3067\u3042\u308a\u3001\u691c\u8a3c\u3055\u308c\u305f\u8a3c\u62e0\u3067\u5224\u65ad\u3092\u4e0b\u3059\u305f\u3081\u306b\u5916\u90e8\u30c4\u30fc\u30eb\u3092\u81ea\u5f8b\u7684\u306b\u547c\u3073\u51fa\u3059\u3002 ARM-Thinker\u3092\u591a\u6bb5\u968e\u5f37\u5316\u5b66\u7fd2\u3067\u8a13\u7df4\u3057\u3001\u30c4\u30fc\u30eb\u30b3\u30fc\u30eb\u6c7a\u5b9a\u3068\u5224\u5b9a\u7cbe\u5ea6\u3092\u5354\u8abf\u7684\u306b\u6700\u9069\u5316\u3059\u308b\u3002 \u305d\u306e\u7d50\u679c,\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u80fd\u529b\u306f\u5831\u916c\u30e2\u30c7\u30eb\u306e\u7cbe\u5ea6\u3068\u89e3\u91c8\u53ef\u80fd\u6027\u306e\u4e21\u65b9\u3092\u8457\u3057\u304f\u5411\u4e0a\u3055\u305b\u308b\u3053\u3068\u304c\u308f\u304b\u3063\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2512.05111v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2512.05111v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 04 Dec 2025 18:59:52 GMT)<\/li>\n\n\n\n<li>\u300cWe introduce ARM-Thinker, an agentic reasoning reward model that judges with an explicit think\u2013 act\u2013verify loop: it plans reasoning steps, invokes multimodal tools (e g , document retrieval and navigation for long PDFs) to gather evidence, and issues an evidencegrounded scalar score with an interpretable rationale.\u300d\u3068Agentic\u306a\u52d5\u4f5c\u3092\u6301\u3064\u624b\u6cd5\u306e\u63d0\u6848\u3002\u300cWe present a scalable data- generation pipeline that constructs verifiable discriminative preference pairs for training agentic reward models. Trained on this data, our ARMThinker-7B achieves performance competitive with, and in some cases superior to, proprietary models like GPT-4o on reward-modeling and tool-use benchmarks, demonstrating the effectiveness of agentic judgment.\u300d\u3068Post training\u306f\u307e\u3060\u307e\u3060\u6709\u529b\u306a\u9078\u629e\u80a2\u306b\u3082\u601d\u3048\u308b\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/InternLM\/ARM-Thinker\">GitHub &#8211; InternLM\/ARM-Thinker: Official Code for &#8220;ARM-Thinker: Reinforcing Multimodal Generative Reward Models with Agentic Tool Use and Visual Reasoning&#8221;<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-7868","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7868","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=7868"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7868\/revisions"}],"predecessor-version":[{"id":7869,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7868\/revisions\/7869"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7868"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7868"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7868"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}