{"id":7139,"date":"2025-07-23T06:28:00","date_gmt":"2025-07-22T21:28:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=7139"},"modified":"2025-07-20T15:31:01","modified_gmt":"2025-07-20T06:31:01","slug":"scaling-up-rl-unlocking-diverse-reasoning-in-llms-via-prolonged-training","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=7139","title":{"rendered":"Scaling Up RL: Unlocking Diverse Reasoning in LLMs via Prolonged Training\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Scaling Up RL: Unlocking Diverse Reasoning in LLMs via Prolonged Training\u00a0<\/strong>[121.6]<br>\u672c\u7814\u7a76\u3067\u306f,\u9577\u671f\u5f37\u5316\u5b66\u7fd2\u304c\u591a\u7a2e\u591a\u69d8\u306a\u63a8\u8ad6\u9818\u57df\u306b\u307e\u305f\u304c\u308b\u5c0f\u8a00\u8a9e\u30e2\u30c7\u30eb\u306b\u53ca\u307c\u3059\u5f71\u97ff\u306b\u3064\u3044\u3066\u691c\u8a0e\u3059\u308b\u3002 \u6211\u3005\u306f,\u9577\u671f\u7684\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u5411\u4e0a\u306e\u9375\u3068\u306a\u308b\u91cd\u8981\u306a\u8981\u7d20\u3068\u3057\u3066,\u5236\u5fa1KL\u6b63\u898f\u5316,\u30af\u30ea\u30c3\u30d4\u30f3\u30b0\u7387,\u5b9a\u671f\u53c2\u7167\u30dd\u30ea\u30b7\u30fc\u30ea\u30bb\u30c3\u30c8\u3092\u5c0e\u5165\u3059\u308b\u3002 \u79c1\u305f\u3061\u306e\u30e2\u30c7\u30eb\u306f\u3001\u6570\u5b66\u306e+14.7%\u3001\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u306e+13.9%\u3001\u8ad6\u7406\u30d1\u30ba\u30eb\u306e+54.8%\u306a\u3069\u3001\u5f37\u529b\u306a\u30d9\u30fc\u30b9\u30e9\u30a4\u30f3\u3088\u308a\u3082\u5927\u5e45\u306b\u6539\u5584\u3055\u308c\u3066\u3044\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2507.12507v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2507.12507v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Wed, 16 Jul 2025 17:59:24 GMT)<\/li>\n\n\n\n<li>\u300cOur work demonstrates that through careful algorithm design, including decoupled clipping, dynamic sampling, controlled KL regularization, and periodic reference policy resets, even small- scale models can achieve substantial reasoning improvements without the computational demands of larger architectures.\u300d\u3068\u5c0f\u898f\u6a21\u30e2\u30c7\u30eb\u3067\u3082\u6709\u52b9\u306a\u5f37\u5316\u5b66\u7fd2\u624b\u6cd5\u306e\u63d0\u6848\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/huggingface.co\/nvidia\/Nemotron-Research-Reasoning-Qwen-1.5B\">nvidia\/Nemotron-Research-Reasoning-Qwen-1.5B \u00b7 Hugging Face<\/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":[576],"class_list":["post-7139","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-576"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7139","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=7139"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7139\/revisions"}],"predecessor-version":[{"id":7140,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7139\/revisions\/7140"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7139"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7139"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7139"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}