{"id":5395,"date":"2024-09-03T06:10:00","date_gmt":"2024-09-02T21:10:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=5395"},"modified":"2024-09-03T06:10:00","modified_gmt":"2024-09-02T21:10:00","slug":"smaller-weaker-yet-better-training-llm-reasoners-via-compute-optimal-sampling","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=5395","title":{"rendered":"Smaller, Weaker, Yet Better: Training LLM Reasoners via Compute-Optimal Sampling\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Smaller, Weaker, Yet Better: Training LLM Reasoners via Compute-Optimal Sampling\u00a0<\/strong>[18.2]<br>\u5f37\u529b\u306a\u8a00\u8a9e\u30e2\u30c7\u30eb(LM)\u3092\u7528\u3044\u305f\u9ad8\u54c1\u8cea\u306a\u5408\u6210\u30c7\u30fc\u30bf\u306e\u5b66\u7fd2\u306f\u3001LM\u306e\u63a8\u8ad6\u6027\u80fd\u3092\u5411\u4e0a\u3055\u305b\u308b\u305f\u3081\u306e\u4e00\u822c\u7684\u306a\u6226\u7565\u3067\u3042\u308b\u3002 \u3088\u308a\u5f37\u529b\u306aSE\u30e2\u30c7\u30eb\u3068\u5f31\u3044\u304c\u5b89\u4fa1\u306aWC\u30e2\u30c7\u30eb\u306b\u3088\u308b\u5408\u6210\u30c7\u30fc\u30bf\u751f\u6210\u306e\u30c8\u30ec\u30fc\u30c9\u30aa\u30d5\u306b\u3064\u3044\u3066\u691c\u8a0e\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2408.16737v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2408.16737v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 29 Aug 2024 17:32:35 GMT)<\/li>\n\n\n\n<li>\u5408\u6210\u30c7\u30fc\u30bf\u751f\u6210\u306b\u304a\u3051\u308bstronger but more expensive (SE) model \u3068 a weaker but cheaper (WC) model\u306e\u6bd4\u8f03\u3002\u300cOur results indicate that it is more compute-optimal to sample from a WC model as opposed to the common-practice of sampling from a SE model.\u300d\u3068\u306e\u3053\u3068\u3002<\/li>\n\n\n\n<li>\u300c3) a new paradigm we introduce called Weak-to-Strong Improvement, where a strong student LM improves using synthetic data from a weaker teacher LM.\u300d\u3068\u3044\u3046\u8a2d\u5b9a\u3001\u304a\u3088\u3073\u3001\u610f\u5916\u306a\u3053\u3068\u306b\u3053\u308c\u304c\u6709\u52b9\u3067\u3042\u308b\u70b9\u3082\u8208\u5473\u6df1\u3044\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":[365,390],"class_list":["post-5395","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-slm","tag-synthetic-data"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5395","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=5395"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5395\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5395"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5395"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5395"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}