{"id":4671,"date":"2024-04-12T05:03:00","date_gmt":"2024-04-11T20:03:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=4671"},"modified":"2024-04-12T05:03:00","modified_gmt":"2024-04-11T20:03:00","slug":"training-llms-over-neurally-compressed-text","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=4671","title":{"rendered":"Training LLMs over Neurally Compressed Text"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Training LLMs over Neurally Compressed Text\u00a0<\/strong>[55.1]<br>\u672c\u7a3f\u3067\u306f,\u9ad8\u5ea6\u306b\u5727\u7e2e\u3055\u308c\u305f\u30c6\u30ad\u30b9\u30c8\u4e0a\u3067\u306e\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u306e\u8a13\u7df4\u306b\u3064\u3044\u3066\u691c\u8a0e\u3059\u308b\u3002 \u30c6\u30ad\u30b9\u30c8\u3092\u30d6\u30ed\u30c3\u30af\u306b\u5206\u5272\u3057,\u305d\u308c\u305e\u308c\u304c\u540c\u3058\u30d3\u30c3\u30c8\u9577\u306b\u5727\u7e2e\u3059\u308b\u65b0\u3057\u3044\u5727\u7e2e\u624b\u6cd5\u3067\u3042\u308bEqual-Info Windows\u3092\u63d0\u6848\u3059\u308b\u3002 \u63d0\u6848\u624b\u6cd5\u306f, \u5927\u898f\u6a21\u5316\u306b\u3088\u308a\u5411\u4e0a\u3057, \u30d1\u30fc\u30d7\u30ec\u30ad\u30b7\u30c6\u30a3\u3068\u63a8\u8ad6\u901f\u5ea6\u306e\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306b\u304a\u3044\u3066, \u30d0\u30a4\u30c8\u30ec\u30d9\u30eb\u306e\u30d9\u30fc\u30b9\u30e9\u30a4\u30f3\u3092\u306f\u308b\u304b\u306b\u4e0a\u56de\u308b, \u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306b\u3088\u308b\u52b9\u679c\u7684\u306a\u5b66\u7fd2\u3092\u5b9f\u6f14\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2404.03626v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2404.03626v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 04 Apr 2024 17:48:28 GMT)<\/li>\n\n\n\n<li>\u5727\u7e2e\u3057\u305f\u30c6\u30ad\u30b9\u30c8\u3092\u7528\u3044\u305f\u5b66\u7fd2\u3001\u300cIn particular, we find that text na\u00efvely compressed via Arithmetic Coding is not readily learnable by LLMs.To overcome this, we propose Equal-Info Windows, a novel compression technique whereby text is segmented into blocks that each compress to the same bit length.\u300d\u3068\u306e\u3053\u3068\u3002<\/li>\n\n\n\n<li>\u3081\u3063\u3061\u3083\u9762\u767d\u3044\u30a2\u30a4\u30c7\u30a2\u3067\u306f\u3042\u308b\u304c\u5b9f\u7528\u7684\u304b\u306f\u8b0e\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":[223,565],"class_list":["post-4671","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm","tag-565"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4671","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=4671"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4671\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4671"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4671"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4671"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}