{"id":5634,"date":"2024-10-25T05:52:00","date_gmt":"2024-10-24T20:52:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=5634"},"modified":"2024-10-25T05:52:00","modified_gmt":"2024-10-24T20:52:00","slug":"merge-to-learn-efficiently-adding-skills-to-language-models-with-model-merging","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=5634","title":{"rendered":"Merge to Learn: Efficiently Adding Skills to Language Models with Model Merging"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Merge to Learn: Efficiently Adding Skills to Language Models with Model Merging\u00a0<\/strong>[102.2]<br>\u6c4e\u7528\u8a00\u8a9e\u30e2\u30c7\u30eb\u3092\u65b0\u3057\u3044\u30b9\u30ad\u30eb\u306b\u9069\u7528\u3059\u308b\u3053\u3068\u306f\u3001\u73fe\u5728\u3001\u9ad8\u4fa1\u306a\u30d7\u30ed\u30bb\u30b9\u3067\u3042\u308b\u3002 \u65e2\u5b58\u306e\u30e2\u30c7\u30eb\u306b\u65b0\u305f\u306a\u30b9\u30ad\u30eb\u3092\u4ed8\u52a0\u3059\u308b\u52b9\u679c\u306b\u3064\u3044\u3066,\u65b0\u305f\u306a\u30b9\u30ad\u30eb\u3092\u5358\u72ec\u3067\u8a13\u7df4\u3057,\u305d\u306e\u5f8c\u4e00\u822c\u30e2\u30c7\u30eb\u3068\u30de\u30fc\u30b8\u3059\u308b\u3053\u3068\u306b\u3088\u3063\u3066\u691c\u8a0e\u3057\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2410.12937v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2410.12937v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Wed, 16 Oct 2024 18:23:50 GMT)<\/li>\n\n\n\n<li>\u300cAs training datasets targeting new skills are constructed, it is an open question how best to patch preexisting models to incorporate the new skills represented by those datasets.\u300d\u3068\u3044\u3046\u72b6\u6cc1\u3067\u306e\u300ccontinued finetuning (CFT) \u300d\u3001\u300cretraining (RT)\u300d\u3001\u300cparallel train then merge (PTM)\u300d\u306e\u6bd4\u8f03<\/li>\n\n\n\n<li>\u300cWe find that PTM is an efficient and effective method of augmenting preexisting models, enabling the addition of new skills with a fraction of the compute required compared to other common methods.\u300d\u3068\u7d50\u8ad6<\/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":[256],"class_list":["post-5634","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-model-merging"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5634","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=5634"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5634\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5634"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5634"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5634"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}