{"id":4503,"date":"2024-03-04T05:05:00","date_gmt":"2024-03-03T20:05:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=4503"},"modified":"2024-03-04T05:05:00","modified_gmt":"2024-03-03T20:05:00","slug":"1-bit1-58-bit%e3%81%aallm%e3%81%a8hawk%e3%83%bbgriffin","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=4503","title":{"rendered":"1 bit(1.58 bit)\u306aLLM\u3068HAWK\u30fbGriffin"},"content":{"rendered":"\n<p>LLM\u306b\u304a\u3044\u3066\u9ad8\u901f\u5316\u306f\u975e\u5e38\u306b\u91cd\u8981\u3002\u5148\u9031\u30d0\u30ba\u3063\u305f\u300cThe Era of 1-bit LLMs\u300d\u3068\u300cGriffin\u300d\u306f\u7570\u306a\u308b\u30a2\u30d7\u30ed\u30fc\u30c1\u3067\u306f\u3042\u308b\u304c\u4eca\u306e\u6a19\u6e96\u30ec\u30b7\u30d4\u3092\u6539\u5584\u3057\u3066\u3044\u304f\u53d6\u308a\u7d44\u307f\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits\u00a0<\/strong>[129.7]<br>\u6211\u3005\u306f1\u30d3\u30c3\u30c8\u306eLarge Language Models (LLMs) \u3001\u3059\u306a\u308f\u3061BitNet b1.58\u3092\u5c0e\u5165\u3059\u308b\u3002 1.58\u30d3\u30c3\u30c8 LLM \u306f\u3001\u65b0\u3057\u3044\u4e16\u4ee3\u306e LLM \u3092\u8a13\u7df4\u3059\u308b\u305f\u3081\u306e\u65b0\u3057\u3044\u30b9\u30b1\u30fc\u30ea\u30f3\u30b0\u6cd5\u5247\u3068\u30ec\u30b7\u30d4\u3092\u5b9a\u7fa9\u3057\u3066\u3044\u308b\u3002 \u3053\u308c\u306f\u65b0\u3057\u3044\u30d1\u30e9\u30c0\u30a4\u30e0\u3092\u53ef\u80fd\u306b\u3057\u30011\u30d3\u30c3\u30c8LLM\u5411\u3051\u306b\u6700\u9069\u5316\u3055\u308c\u305f\u7279\u5b9a\u306e\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u3092\u8a2d\u8a08\u3059\u308b\u305f\u3081\u306e\u6249\u3092\u958b\u304f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2402.17764v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2402.17764v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Tue, 27 Feb 2024 18:56:19 GMT)<\/li>\n\n\n\n<li>-1, 0, 1\u306e\u307f\u3092\u7528\u3044\u305fLLM\uff08\u5b66\u7fd2\u6642\u306f8bit\uff09\u3001\u9ad8\u901f\u306a\u5b66\u7fd2\u30fb\u63a8\u8ad6\u304c\u53ef\u80fd<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/microsoft\/unilm?tab=readme-ov-file#the-revolution-of-model-architecture\">microsoft\/unilm: Large-scale Self-supervised Pre-training Across Tasks, Languages, and Modalities (github.com)<\/a>\u3001\u975e\u516c\u5f0f\u306a\u5b9f\u88c5\u304c\u3042\u308b\u3088\u3046<a href=\"https:\/\/github.com\/kyegomez\/BitNet\">kyegomez\/BitNet: Implementation of &#8220;BitNet: Scaling 1-bit Transformers for Large Language Models&#8221; in pytorch (github.com)<\/a><\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Griffin: Mixing Gated Linear Recurrences with Local Attention for Efficient Language Models\u00a0<\/strong>[101.7]<br>\u6211\u3005\u306f\u3001\u30b2\u30fc\u30c8\u7dda\u5f62\u518d\u5e30\u3092\u6301\u3064RNN\u306eHawk\u3068\u3001\u30b2\u30fc\u30c8\u7dda\u5f62\u518d\u5e30\u3068\u5c40\u6240\u7684\u306a\u6ce8\u610f\u3092\u6df7\u5408\u3057\u305f\u30cf\u30a4\u30d6\u30ea\u30c3\u30c9\u30e2\u30c7\u30ebGriffin\u3092\u63d0\u6848\u3059\u308b\u3002 \u30db\u30fc\u30af\u306f\u4e0b\u6d41\u3067\u306e\u30de\u30f3\u30d0\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u4e0a\u56de\u308a\u3001\u30b0\u30ea\u30d5\u30a3\u30f3\u306f6\u500d\u4ee5\u4e0a\u306e\u30c8\u30fc\u30af\u30f3\u3067\u8a13\u7df4\u3055\u308c\u3066\u3044\u308b\u306b\u3082\u304b\u304b\u308f\u3089\u305a\u3001\u30e9\u30de-2\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u306b\u5339\u6575\u3059\u308b\u3002 Griffin\u309214B\u30d1\u30e9\u30e1\u30fc\u30bf\u307e\u3067\u62e1\u5f35\u3057\u3001\u52b9\u7387\u7684\u306a\u5206\u6563\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u306e\u305f\u3081\u306b\u30e2\u30c7\u30eb\u3092\u30b7\u30e3\u30fc\u30c7\u30a3\u30f3\u30b0\u3059\u308b\u65b9\u6cd5\u3092\u8aac\u660e\u3057\u307e\u3059\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2402.19427v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2402.19427v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 29 Feb 2024 18:24:46 GMT)<\/li>\n\n\n\n<li>RNN\u30d9\u30fc\u30b9\u306eHAWK\u3001\u30cf\u30a4\u30d6\u30ea\u30c3\u30c9\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306eGriffin\u306e\u63d0\u6848\u3002HAWK\u306f\u975e\u5e38\u306b\u9ad8\u901f\u3060\u304c\u3001\u6027\u80fd\u306f\u540c\u898f\u6a21\u306eTransformer\u3068\u7af6\u5408\u3059\u308b\u30ec\u30d9\u30eb\uff08Mamba\u3088\u308a\u512a\u308c\u3066\u3044\u308b\u3088\u3046\uff09\u3002Griffin\u306f\u305d\u308c\u3092outperform\u3057\u3066\u3044\u308b\u3088\u3046\u306b\u898b\u3048\u308b\u30027B\u308414B\u3068\u5927\u898f\u6a21\u306a\u691c\u8a3c\u3092\u3057\u3066\u3044\u308b\u3042\u305f\u308a\u3055\u3059\u304cGoogle DeepMind\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>LLM\u306b\u304a\u3044\u3066\u9ad8\u901f\u5316\u306f\u975e\u5e38\u306b\u91cd\u8981\u3002\u5148\u9031\u30d0\u30ba\u3063\u305f\u300cThe Era of 1-bit LLMs\u300d\u3068\u300cGriffin\u300d\u306f\u7570\u306a\u308b\u30a2\u30d7\u30ed\u30fc\u30c1\u3067\u306f\u3042\u308b\u304c\u4eca\u306e\u6a19\u6e96\u30ec\u30b7\u30d4\u3092\u6539\u5584\u3057\u3066\u3044\u304f\u53d6\u308a\u7d44\u307f\u3002<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[223,415],"class_list":["post-4503","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm","tag-transformer"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4503","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=4503"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4503\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4503"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4503"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4503"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}