{"id":7126,"date":"2025-07-24T05:29:00","date_gmt":"2025-07-23T20:29:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=7126"},"modified":"2025-07-20T14:35:54","modified_gmt":"2025-07-20T05:35:54","slug":"lizard-an-efficient-linearization-framework-for-large-language-models","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=7126","title":{"rendered":"Lizard: An Efficient Linearization Framework for Large Language Models"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Lizard: An Efficient Linearization Framework for Large Language Models\u00a0<\/strong>[100.6]<br>\u6211\u3005\u306f,\u4e8b\u524d\u5b66\u7fd2\u3057\u305f\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30d9\u30fc\u30b9\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u3092,\u7121\u9650\u30b3\u30f3\u30c6\u30ad\u30b9\u30c8\u751f\u6210\u306e\u305f\u3081\u306e\u67d4\u8edf\u6027\u306e\u3042\u308b\u30b5\u30d6\u30af\u30ef\u30c3\u30c9\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306b\u5909\u63db\u3059\u308b\u7dda\u5f62\u5316\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3042\u308bLizard\u3092\u63d0\u6848\u3059\u308b\u3002 Lizard\u306f\u3001\u51fa\u529b\u54c1\u8cea\u3092\u4fdd\u3061\u306a\u304c\u3089\u30bd\u30d5\u30c8\u30de\u30c3\u30af\u30b9\u30a2\u30c6\u30f3\u30b7\u30e7\u30f3\u3092\u6b63\u78ba\u306b\u8fd1\u4f3c\u3059\u308b\u30b5\u30d6\u30af\u30ef\u30c3\u30c9\u30a2\u30c6\u30f3\u30b7\u30e7\u30f3\u30e1\u30ab\u30cb\u30ba\u30e0\u3092\u5c0e\u5165\u3059\u308b\u3053\u3068\u3067\u3001\u3053\u306e\u5236\u9650\u306b\u5bfe\u51e6\u3059\u308b\u3002 \u305d\u3053\u3067\u672c\u7814\u7a76\u3067\u306f,Lizard\u304c\u5f93\u6765\u306e\u7dda\u5f62\u5316\u624b\u6cd5\u3092\u8457\u3057\u304f\u4e0a\u56de\u308a\u306a\u304c\u3089,\u6a19\u6e96\u8a00\u8a9e\u30e2\u30c7\u30ea\u30f3\u30b0\u30bf\u30b9\u30af\u306b\u304a\u3051\u308b\u6559\u5e2b\u30e2\u30c7\u30eb\u306e\u6027\u80fd\u306e\u307b\u307c\u7121\u4f5c\u70ba\u306a\u56de\u5fa9\u3092\u5b9f\u73fe\u3057\u3066\u3044\u308b\u3053\u3068\u3092\u793a\u3059\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2507.09025v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2507.09025v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Fri, 11 Jul 2025 21:19:18 GMT)<\/li>\n\n\n\n<li>\u300cLizard (Linearizing Softmax Attention with Recurrent Gate Dynamics), an efficient framework for linearizing LLMs\u300d\u306e\u63d0\u6848\u3002<\/li>\n\n\n\n<li>\u300cWe train our model in two stages: (1) an attention approximation stage where the subquadratic modules are trained to mimic softmax attention outputs, and (2) a fine-tuning stage where the linearized model is adapted to downstream language modeling objectives.\u300d\u3068\u65e2\u5b58\u30e2\u30c7\u30eb\u3092\u6d3b\u7528\u3057\u3066\u3044\u304f\u30a2\u30d7\u30ed\u30fc\u30c1\u3067\u62e1\u5f35\u306b\u4f7f\u7528\u3059\u308b\u5b66\u7fd2\u30c7\u30fc\u30bf\u304c\u5c11\u306a\u304f\u3001\u6027\u80fd\u52a3\u5316\u3082\u6291\u3048\u3089\u308c\u308b\u3068\u306e\u3053\u3068\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":[33,415],"class_list":["post-7126","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-attention","tag-transformer"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7126","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=7126"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7126\/revisions"}],"predecessor-version":[{"id":7127,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7126\/revisions\/7127"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7126"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7126"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7126"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}