{"id":5752,"date":"2024-11-11T06:16:00","date_gmt":"2024-11-10T21:16:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=5752"},"modified":"2024-11-11T06:16:00","modified_gmt":"2024-11-10T21:16:00","slug":"number-cookbook-number-understanding-of-language-models-and-how-to-improve-it","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=5752","title":{"rendered":"Number Cookbook: Number Understanding of Language Models and How to Improve It\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Number Cookbook: Number Understanding of Language Models and How to Improve It\u00a0<\/strong>[64.0]<br>\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u306f\u3001\u57fa\u672c\u7684\u306a\u6570\u5024\u7684\u306a\u7406\u89e3\u3068\u51e6\u7406\u306b\u304a\u3044\u3066\u4e88\u671f\u305b\u306c\u8aa4\u308a\u3092\u72af\u3057\u306a\u304c\u3089\u3001\u8907\u96d1\u306a\u63a8\u8ad6\u30bf\u30b9\u30af\u306e\u5897\u5927\u3092\u89e3\u6c7a\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u308b\u3002 \u672c\u7a3f\u3067\u306f,LLM\u306e\u6570\u5024\u7406\u89e3\u3068\u51e6\u7406\u80fd\u529b(NUPA)\u306b\u3064\u3044\u3066\u5305\u62ec\u7684\u306b\u691c\u8a0e\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2411.03766v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2411.03766v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Wed, 06 Nov 2024 08:59:44 GMT)<\/li>\n\n\n\n<li>LLM\u306b\u304a\u3051\u308b\u00a0numerical understanding and processing ability (NUPA)\u306e\u5206\u6790\u3068\u3001\u305d\u306e\u6539\u5584\u65b9\u6cd5\u306e\u691c\u8a0e\u3002\u73fe\u72b6\u3060\u3068\u30b3\u30fc\u30c9\u751f\u6210\u3092\u4ecb\u3059\u306a\u3069\u30c4\u30fc\u30eb\u3092\u4f7f\u3046\u30a2\u30d7\u30ed\u30fc\u30c1\u304c\u6709\u529b\u3060\u304c\u3001\u300c1) we want to study the self-contained NUPA of LLMs,\u30002) calling external tools whenever encountering numbers increases the inference latency (Xu et al , 2024), and 3) we believe NUPA without tools is a necessary ability of AGI.\u300d\u3068\u3044\u3046\u70b9\u304b\u3089\u672c\u4ef6\u3067\u306f\u30c4\u30fc\u30eb\u5229\u7528\u304c\u691c\u8a0e\u5bfe\u8c61\u5916\u3068\u306a\u3063\u3066\u3044\u308b\u3002<\/li>\n\n\n\n<li>\u73fe\u6642\u70b9\u3067\u306f\u300cWe investigate NUPA of LLMs and introduce a comprehensive benchmark, the NUPA test, to reveal that numerical problems remain challenging for modern LLMs.\u300d\u3068\u306e\u3053\u3068\u3002\u3084\u306f\u308a\u96e3\u3057\u3044\u554f\u984c\u3002\u5b9f\u7528\u4e0a\u306f\u30b3\u30fc\u30c9\u751f\u6210\u3092\u4ecb\u3059\u306a\u3069\u3057\u3066\u5bfe\u5fdc\u3067\u304d\u306a\u304f\u306f\u306a\u3044\u304c\u30fb\u30fb\u30fb\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/GraphPKU\/number_cookbook\">GitHub &#8211; GraphPKU\/number_cookbook<\/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":[223,588,589],"class_list":["post-5752","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm","tag-588","tag-589"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5752","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=5752"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5752\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5752"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5752"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5752"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}