{"id":4826,"date":"2024-05-06T07:08:00","date_gmt":"2024-05-05T22:08:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=4826"},"modified":"2024-05-06T07:08:00","modified_gmt":"2024-05-05T22:08:00","slug":"a-careful-examination-of-large-language-model-performance-on-grade-school-arithmetic","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=4826","title":{"rendered":"A Careful Examination of Large Language Model Performance on Grade School Arithmetic\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>A Careful Examination of Large Language Model Performance on Grade School Arithmetic\u00a0<\/strong>[4.7]<br>\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb (LLM) \u306f\u3001\u6570\u5b66\u7684\u63a8\u8ad6\u306e\u305f\u3081\u306e\u591a\u304f\u306e\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3067\u9a5a\u304f\u3079\u304d\u6210\u529f\u3092\u53ce\u3081\u305f\u3002 \u3053\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u306e\u4e00\u90e8\u306f\u3001\u5b9f\u969b\u306b\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306e\u6c5a\u67d3\u3092\u53cd\u6620\u3057\u3066\u3044\u308b\u3001\u3068\u3044\u3046\u61f8\u5ff5\u304c\u9ad8\u307e\u3063\u3066\u3044\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2405.00332v2\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2405.00332v2\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 02 May 2024 17:18:51 GMT)<\/li>\n\n\n\n<li>\u300cFigure 1: Notable models arranged by their drop in performance between GSM8k and GSM1k (lower is worse).We notice that Mistral and Phi top the list of overfit models, with almost 10% drops on GSM1k compared to GSM8k, while models such as Gemini, GPT, and Claude show little to no signs of overfitting.\u300d\u304c\u885d\u6483\u7684\u306a\u8ad6\u6587\u3067\u3001\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u30c7\u30fc\u30bf\u306e\u6f5c\u5728\u7684\u306aLeak\u304c\u554f\u984c\u306b\u306a\u3063\u3066\u3044\u308b\u3053\u3068\u3092\u793a\u3057\u3066\u3044\u308b\u3002<\/li>\n\n\n\n<li><a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2309.08632v1\">Fugu-MT \u8ad6\u6587\u7ffb\u8a33(\u6982\u8981): Pretraining on the Test Set Is All You Need (fugumt.com)<\/a>\u3084<a href=\"https:\/\/note.com\/kan_hatakeyama\/n\/n5a85700bf46c\">\u5546\u7528\u5229\u7528\u53ef\u80fd\u306a130\u5104\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u65e5\u672c\u8a9eLLM\u300cTanuki-ZeRo\u300d\u3092\u4e00\u822c\u516c\u958b \u3010\u4ee3\u8868\u7684\u306a\u65e5\u672c\u8a9e\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3067\u4e16\u754c6\u4f4d: \u30aa\u30fc\u30d7\u30f3\u30e2\u30c7\u30eb\u30671\u4f4d\u76f8\u5f53\uff64GPT3.5\u3084Claude v2\u3092\u4e00\u90e8\u51cc\u99d5, 23\uff5cKan Hatakeyama (note.com)<\/a>\u306a\u3069\u610f\u56f3\u7684\u306b\u826f\u304f\u3082\u3067\u304d\u308b\u304c\u3001\u305d\u3046\u3067\u306a\u304f\u3068\u3082\u6839\u6df1\u3044\u554f\u984c\u3067\u5bfe\u7b56\u306f\u7c21\u5358\u3067\u306f\u306a\u3044\u3002<\/li>\n\n\n\n<li>\u500b\u4eba\u306e\u691c\u8a3c\u3067\u3082Phi-3\u3082\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u7d50\u679c\u7a0b\u3088\u304f\u306f\u306a\u3044\u306e\u3067\u306f\u306a\u3044\u304b\u3068\u601d\u3046\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":[216,223],"class_list":["post-4826","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-leakage","tag-llm"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4826","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=4826"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4826\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4826"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4826"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4826"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}