{"id":6684,"date":"2025-05-05T04:43:00","date_gmt":"2025-05-04T19:43:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6684"},"modified":"2025-05-05T04:43:00","modified_gmt":"2025-05-04T19:43:00","slug":"reinforcement-learning-for-reasoning-in-large-language-models-with-one-training-example","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6684","title":{"rendered":"Reinforcement Learning for Reasoning in Large Language Models with One Training Example"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Reinforcement Learning for Reasoning in Large Language Models with One Training Example\u00a0<\/strong>[129.1]<br>1\u3064\u306e\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u4f8b(1\u30b7\u30e7\u30c3\u30c8RLVR)\u3092\u7528\u3044\u305f\u5f37\u5316\u5b66\u7fd2\u306f,\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u306e\u7b97\u6570\u63a8\u8ad6\u80fd\u529b\u306e\u5411\u4e0a\u306b\u6709\u52b9\u3067\u3042\u308b\u3053\u3068\u3092\u793a\u3059\u3002 1\u30b7\u30e7\u30c3\u30c8RLVR\u306b\u304a\u3051\u308b\u8208\u5473\u6df1\u3044\u73fe\u8c61\u3068\u3057\u3066\u3001\u30af\u30ed\u30b9\u30c9\u30e1\u30a4\u30f3\u306e\u4e00\u822c\u5316\u3001\u81ea\u5df1\u56de\u5e30\u306e\u983b\u5ea6\u306e\u5897\u5927\u3001\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u7cbe\u5ea6\u304c\u98fd\u548c\u3057\u305f\u5f8c\u3082\u30c6\u30b9\u30c8\u6027\u80fd\u306e\u5411\u4e0a\u304c\u7dad\u6301\u3055\u308c\u3066\u3044\u308b\u3053\u3068\u3092\u6319\u3052\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2504.20571v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2504.20571v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Tue, 29 Apr 2025 09:24:30 GMT)<\/li>\n\n\n\n<li>\u300cWe find that selecting one specific example as the training dataset can achieve similar downstream performance to that of the 1.2k DeepScaleR subset (DSR-sub) containing that example. Specifically, this improves the Qwen2.5-Math-1.5B model from 36.0% to 73.6% on MATH500, and from 17.6% to 35.7% on average across 6 mathematical reasoning benchmarks (Fig. 1, 2).\u300d\u3068\u3044\u3046\u8208\u5473\u6df1\u3044\u5831\u544a\u3002\u300cThese findings suggest that the reasoning capability of the model is already buried in the base model, and encouraging exploration on a very small amount of data is capable of generating useful RL training signals for igniting LLM\u2019s reasoning capability.\u300d\u306f\u305d\u3046\u306a\u306e\u3060\u308d\u3046\u3068\u601d\u3046\u3002LLM\u306e\u4e2d\u306b\u306f\u4f55\u304c\u5165\u3063\u3066\u3044\u3066\u30c1\u30e5\u30fc\u30cb\u30f3\u30b0\u3063\u3066\u4f55\u3092\u3057\u3066\u3044\u308b\u3093\u3060\u308d\u3046\u3002\u3002\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/ypwang61\/One-Shot-RLVR\">GitHub &#8211; ypwang61\/One-Shot-RLVR: official repository for \u201cReinforcement Learning for Reasoning in Large Language Models with One Training Example\u201d<\/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":[232,366],"class_list":["post-6684","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\/6684","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=6684"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6684\/revisions"}],"predecessor-version":[{"id":6699,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6684\/revisions\/6699"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6684"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6684"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6684"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}