{"id":7260,"date":"2025-08-14T05:37:00","date_gmt":"2025-08-13T20:37:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=7260"},"modified":"2025-08-11T13:42:48","modified_gmt":"2025-08-11T04:42:48","slug":"mlp-memory-language-modeling-with-retriever-pretrained-external-memory","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=7260","title":{"rendered":"MLP Memory: Language Modeling with Retriever-pretrained External Memory\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>MLP Memory: Language Modeling with Retriever-pretrained External Memory\u00a0<\/strong>[26.0]<br>\u305d\u3053\u3067\u672c\u7814\u7a76\u3067\u306f,\u4e8b\u524d\u5b66\u7fd2\u53ef\u80fd\u306a\u5916\u90e8\u30e1\u30e2\u30ea\u3092\u7528\u3044\u3066\u30c7\u30b3\u30fc\u30c0\u304b\u3089\u5207\u308a\u96e2\u3059\u3053\u3068\u3092\u63d0\u6848\u3059\u308b\u3002 \u79c1\u305f\u3061\u306e\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306f\u3001\u4e0b\u6d41\u306e\u30bf\u30b9\u30af\u306b\u5f37\u3044\u96e3\u6613\u5ea6\u3068\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u793a\u3057\u307e\u3059\u3002 3\u3064\u306e\u5e7b\u899a\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u30689\u3064\u306e\u30e1\u30e2\u30ea\u96c6\u7d04\u30bf\u30b9\u30af\u306b\u304a\u3044\u3066\u512a\u308c\u305f\u6027\u80fd\u3092\u793a\u3059\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2508.01832v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2508.01832v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Sun, 03 Aug 2025 16:40:53 GMT)<\/li>\n\n\n\n<li>\u300cIn this work, we propose an external memory for LLM that is pretrained to mimic a retriever on the entire pretraining dataset. Specifically, following the RAG setting in kNN-LM [27], this memory learns to map the LLM hidden state at a certain step to a vocabulary distribution matching the output of the kNN retriever. During inference, the LLM\u2019s native output is interpolated with the retriever-pretrained output from the external memory.\u300d\u3068\u8a18\u61b6\uff08\u77e5\u8b58\uff09\u90e8\u5206\u3092\u5207\u308a\u96e2\u3057\u305f\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306e\u63d0\u6848<\/li>\n\n\n\n<li>\u3053\u308c\u304c\u3046\u307e\u304f\u52d5\u4f5c\u3059\u308b\u306e\u3067\u3042\u308c\u3070\u9762\u767d\u3044\u306a\u3068\u601d\u3046\u4e00\u65b9\u3067\u3001\u77e5\u8b58\u3068\u601d\u8003\u304c\u5207\u308a\u96e2\u305b\u308b\u306e\u304b\u306f\u3084\u3084\u7591\u554f\u3067\u601d\u8003\u30fb\u751f\u6210\u90e8\u5206\u3078\u306e\u5f71\u97ff\u304c\u6c17\u306b\u306a\u308b\u3068\u3053\u308d\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":[244],"class_list":["post-7260","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-memory"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7260","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=7260"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7260\/revisions"}],"predecessor-version":[{"id":7261,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7260\/revisions\/7261"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7260"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7260"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7260"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}