{"id":7066,"date":"2025-07-18T05:33:00","date_gmt":"2025-07-17T20:33:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=7066"},"modified":"2025-07-06T10:52:15","modified_gmt":"2025-07-06T01:52:15","slug":"reasoning-on-a-budget-a-survey-of-adaptive-and-controllable-test-time-compute-in-llms","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=7066","title":{"rendered":"Reasoning on a Budget: A Survey of Adaptive and Controllable Test-Time Compute in LLMs\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Reasoning on a Budget: A Survey of Adaptive and Controllable Test-Time Compute in LLMs&nbsp;<\/strong>[45.8]<br>\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u306f\u3001\u5e45\u5e83\u3044\u30bf\u30b9\u30af\u3092\u89e3\u304f\u3053\u3068\u304c\u3067\u304d\u308b\u6c4e\u7528\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u3078\u3068\u6025\u901f\u306b\u9032\u6b69\u3057\u3066\u304d\u305f\u3002 \u5f7c\u3089\u306f\u3001\u30bf\u30b9\u30af\u306e\u8907\u96d1\u3055\u306b\u95a2\u308f\u3089\u305a\u3001\u56fa\u5b9a\u63a8\u8ad6\u6642\u9593\u8a08\u7b97\u3092\u9069\u7528\u3057\u3001\u3057\u3070\u3057\u3070\u96e3\u3057\u3044\u3053\u3068\u3092\u8003\u3048\u306a\u304c\u3089\u5358\u7d14\u306a\u554f\u984c\u3092\u904e\u5c0f\u8a55\u4fa1\u3059\u308b\u3002 \u672c\u8abf\u67fb\u3067\u306f, LLM\u63a8\u8ad6\u306e\u8a08\u7b97\u52b9\u7387\u5411\u4e0a\u3092\u76ee\u7684\u3068\u3057\u305f, \u52b9\u7387\u7684\u306a\u30c6\u30b9\u30c8\u6642\u9593\u8a08\u7b97\u6226\u7565\u306e\u7dcf\u5408\u7684\u306a\u30ec\u30d3\u30e5\u30fc\u3092\u884c\u3046\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2507.02076v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2507.02076v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Wed, 02 Jul 2025 18:27:42 GMT)<\/li>\n\n\n\n<li>\u300cThis survey presents a comprehensive review of efficient test-time compute (TTC) strategies, which aim to improve the computational efficiency of LLM reasoning. We introduce a two-tiered taxonomy that distinguishes between L1 controllability\u2014methods that operate under fixed compute budgets\u2014and L2 adaptiveness\u2014methods that dynamically scale inference based on input difficulty or model confidence.\u300d\u3068\u3044\u3046\u30b5\u30fc\u30d9\u30a4\u3002<\/li>\n\n\n\n<li>\u5546\u7528\u30e2\u30c7\u30eb\u3067\u306e\u30cf\u30a4\u30d6\u30ea\u30c3\u30c9\u30a2\u30d7\u30ed\u30fc\u30c1\u3082\u6d41\u884c\u3063\u3066\u3044\u3066\u8272\u3005\u3068\u82e6\u52b4\u3057\u3066\u3044\u308b\u90e8\u5206\u306a\u3093\u3060\u308d\u3046\u306a\u3068\u601d\u3046\u3002<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Predicting thinking time in Reasoning models\u00a0<\/strong>[42.6]<br>\u63a8\u8ad6\u30e2\u30c7\u30eb\u306f\u9577\u304f\u96a0\u308c\u305f\u601d\u8003\u306e\u9023\u9396\u3092\u751f\u307f\u51fa\u3057\u307e\u3059\u3002 \u30e6\u30fc\u30b6\u30fc\u306f\u3001\u7b54\u3048\u3092\u8fd4\u3059\u524d\u306b\u30e2\u30c7\u30eb\u304c\u63a8\u8ad6\u306b\u3069\u308c\u304f\u3089\u3044\u306e\u6642\u9593\u3092\u8cbb\u3084\u3059\u304b\u306b\u3064\u3044\u3066\u306e\u6d1e\u5bdf\u304c\u307b\u3068\u3093\u3069\u306a\u3044\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2506.23274v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2506.23274v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Sun, 29 Jun 2025 15:01:01 GMT)<\/li>\n\n\n\n<li>LRM\u306b\u304a\u3051\u308b\u63a8\u8ad6\u6642\u9593\u306e\u4e88\u6e2c\u306b\u95a2\u3059\u308b\u5831\u544a\u3002<\/li>\n\n\n\n<li>\u300cIn this paper, we explore methods for online prediction of thinking time in reasoning models. Our experiments demonstrate that current models encode a notion of progress in their internal representations, with an mlp probe achieving 45% accuracy over 10 classes, moreover the errors appear highly local (MAE 1).\u300d<\/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":[2],"tags":[232,366],"class_list":["post-7066","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-lrm","tag-slow-thinking"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7066","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=7066"}],"version-history":[{"count":2,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7066\/revisions"}],"predecessor-version":[{"id":7071,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7066\/revisions\/7071"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7066"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7066"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7066"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}