{"id":6457,"date":"2025-03-21T05:46:00","date_gmt":"2025-03-20T20:46:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6457"},"modified":"2025-03-21T05:46:00","modified_gmt":"2025-03-20T20:46:00","slug":"light-r1-curriculum-sft-dpo-and-rl-for-long-cot-from-scratch-and-beyond","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6457","title":{"rendered":"Light-R1: Curriculum SFT, DPO and RL for Long COT from Scratch and Beyond"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Light-R1: Curriculum SFT, DPO and RL for Long COT from Scratch and Beyond\u00a0<\/strong>[14.4]<br>\u79c1\u305f\u3061\u306f\u307e\u305a\u3001\u9577\u3044COT\u80fd\u529b\u3092\u6b20\u3044\u305f\u30e2\u30c7\u30eb\u304b\u3089\u59cb\u307e\u308b\u3001\u9577\u3044COT\u30e2\u30c7\u30eb\u3092\u30b9\u30af\u30e9\u30c3\u30c1\u304b\u3089\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3059\u308b\u3053\u3068\u306b\u91cd\u70b9\u3092\u7f6e\u3044\u3066\u3044\u307e\u3059\u3002 Qwen2.5-32B-Instruct\u304b\u30892\u6bb5\u968e\u306eSFT\u3068\u30bb\u30df\u30aa\u30f3\u30fb\u30dd\u30ea\u30c6\u30a3\u30af\u30b9DPO\u304b\u3089\u306a\u308b\u30ab\u30ea\u30ad\u30e5\u30e9\u30e0\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u30ec\u30b7\u30d4\u3092\u7528\u3044\u3066\u3001\u6211\u3005\u306e\u30e2\u30c7\u30eb\u3067\u3042\u308bLight-R1-32B\u3092\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3059\u308b\u3002 AIME24\u306825\u306e\u30b9\u30b3\u30a2\u306f\u305d\u308c\u305e\u308c74.0\u306860.2\u3067\u3042\u308a\u3001Light-R1-14B-DS\u306f32B\u30e2\u30c7\u30eb\u3068DeepSeek-R1-Distill-Llama-70B\u3092\u629c\u3044\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2503.10460v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2503.10460v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 13 Mar 2025 15:29:22 GMT)<\/li>\n\n\n\n<li>2\u30b9\u30c6\u30fc\u30b8\u306eSFT\uff0bDPO Optimization\uff08\uff0b model merge\uff09\u3067\u69cb\u7bc9\u3057\u305f\u30e2\u30c7\u30eb\u3002\u300cHigh-Quality Data is All You Need\u300d\u306e\u901a\u308a\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u5074\u306e\u30d1\u30a4\u30d7\u30e9\u30a4\u30f3\u3082\u51dd\u3063\u3066\u3044\u308b\u3002\u4ed6\u306e\u7814\u7a76\u6210\u679c\u3067\u3082\u8fd1\u3044\u3053\u3068\u304c\u6307\u6458\u3055\u308c\u3066\u3044\u308b\u304c\u300cDespite being trained exclusively on math data, Light-R1-32B shows strong generalization across other domains.\u300d\u306f\u8208\u5473\u6df1\u3044\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/Qihoo360\/Light-R1\">GitHub &#8211; Qihoo360\/Light-R1<\/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":[2],"tags":[283,366],"class_list":["post-6457","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-o1","tag-slow-thinking"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6457","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=6457"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6457\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6457"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6457"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6457"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}