{"id":5532,"date":"2024-10-01T05:13:00","date_gmt":"2024-09-30T20:13:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=5532"},"modified":"2024-10-01T05:13:00","modified_gmt":"2024-09-30T20:13:00","slug":"empirical-insights-on-fine-tuning-large-language-models-for-question-answering","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=5532","title":{"rendered":"Empirical Insights on Fine-Tuning Large Language Models for Question-Answering\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Empirical Insights on Fine-Tuning Large Language Models for Question-Answering\u00a0<\/strong>[50.1]<br>\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u306f\u3001\u5927\u91cf\u306e\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306e\u4e8b\u524d\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3092\u901a\u3058\u3066\u3001\u5e83\u7bc4\u56f2\u306a\u4e16\u754c\u306e\u77e5\u8b58\u3092\u7b26\u53f7\u5316\u3059\u308b\u3002 \u6211\u3005\u306f,\u4e8b\u524d\u5b66\u7fd2\u3057\u305fLLM\u304c\u8a18\u61b6\u3059\u308b\u77e5\u8b58\u306e\u91cf\u306b\u57fa\u3065\u3044\u3066,\u6559\u5e2b\u4ed8\u304d\u5fae\u8abf\u6574(SFT)\u30c7\u30fc\u30bf\u3092\u5206\u985e\u3057\u305f\u3002 \u5b9f\u9a13\u306e\u7d50\u679c,SFT\u306e\u6bb5\u968e\u3067\u306f60\u500b\u306e\u30c7\u30fc\u30bf\u30dd\u30a4\u30f3\u30c8\u304c\u4e8b\u524d\u5b66\u7fd2\u4e2d\u306b\u7b26\u53f7\u5316\u3055\u308c\u305f\u77e5\u8b58\u3092\u6d3b\u6027\u5316\u3059\u308b\u3053\u3068\u304c\u3067\u304d,LLM\u304cQA\u30bf\u30b9\u30af\u3092\u5b9f\u884c\u3067\u304d\u308b\u3053\u3068\u304c\u308f\u304b\u3063\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2409.15825v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2409.15825v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Tue, 24 Sep 2024 07:38:38 GMT)<\/li>\n\n\n\n<li>\u300cTo our surprise, we find that the fine-tuned model neither forgets the relationship among the other classes nor degrades the features to recognize these classes.\u300d\u3001\u300cWhat really hurts the accuracy is the discrepant logit scales between the fine-tuning classes and the other classes, implying that a simple post-processing calibration would bring back the pre-trained model\u2019s capability and at the same time unveil the feature improvement over all classes.\u300d\u3068\u3044\u3046\u6307\u6458\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/OSU-MLB\/Fine-Tuning-Is-Fine-If-Calibrated\">GitHub &#8211; OSU-MLB\/Fine-Tuning-Is-Fine-If-Calibrated<\/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":[150],"class_list":["post-5532","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-fine-tuning"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5532","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=5532"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5532\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5532"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5532"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5532"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}