{"id":6331,"date":"2025-03-05T20:41:24","date_gmt":"2025-03-05T11:41:24","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6331"},"modified":"2025-03-05T20:41:24","modified_gmt":"2025-03-05T11:41:24","slug":"towards-thinking-optimal-scaling-of-test-time-compute-for-llm-reasoning","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6331","title":{"rendered":"Towards Thinking-Optimal Scaling of Test-Time Compute for LLM Reasoning"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Towards Thinking-Optimal Scaling of Test-Time Compute for LLM Reasoning\u00a0<\/strong>[113.5]<br>\u8fd1\u5e74\u306e\u7814\u7a76\u3067\u306f\u3001\u30e2\u30c7\u30eb\u3092\u3088\u308a\u9577\u3044\u601d\u8003\u306e\u9023\u9396(CoTs)\u3092\u901a\u3057\u3066\u8003\u3048\u308b\u6642\u9593\u3092\u5897\u3084\u3059\u3053\u3068\u3067\u3001\u8907\u96d1\u306a\u63a8\u8ad6\u30bf\u30b9\u30af\u306b\u304a\u3044\u3066\u5927\u5e45\u306a\u6539\u5584\u304c\u5f97\u3089\u308c\u308b\u3053\u3068\u304c\u793a\u3055\u308c\u3066\u3044\u308b\u3002 \u3088\u308a\u9577\u3044CoT\u306b\u3088\u308b\u30b9\u30b1\u30fc\u30ea\u30f3\u30b0\u304c\u3001\u7279\u5b9a\u306e\u30c9\u30e1\u30a4\u30f3\u306b\u304a\u3051\u308bLarge Language Model(LLM)\u306e\u63a8\u8ad6\u6027\u80fd\u3092\u640d\u306a\u3046\u304b\u3069\u3046\u304b\u3092\u8003\u5bdf\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2502.18080v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2502.18080v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Tue, 25 Feb 2025 10:48:05 GMT)<\/li>\n\n\n\n<li>\u5341\u5206\u306aCoT\u3092\u63d0\u4f9b\u304b\u3064\u9577\u3059\u304e\u308bCoT\u304c\u60aa\u5f71\u97ff\u3092\u4e0e\u3048\u306a\u3044\u3088\u3046\u306b\u3059\u308b\u300cThinking-OPtimal Scaling strategy (TOPS) that allows LLMs to decide by themselves how many tokens are needed to solve a given problem.\u300d\u306e\u63d0\u6848<\/li>\n\n\n\n<li>\u300cFormat Imitation enables the base model to learn how to adopt different levels of reasoning effort ei to perform System-2 thinking, using a small set of seed data. Reasoning Effort-Conditioned Generation requires the model to apply System-2 thinking to a large set of problems under different reasoning efforts. Self-Improvement select the shortest correct response for each problem among all responses to fine-tune the base model to achieve thinking-optimal test-time scaling.\u300d\u3068\u3044\u30463\u30b9\u30c6\u30fc\u30b8\u69cb\u6210\u3002<\/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":[232,283,366],"class_list":["post-6331","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-lrm","tag-o1","tag-slow-thinking"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6331","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=6331"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6331\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6331"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6331"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6331"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}