{"id":6043,"date":"2025-01-15T05:15:00","date_gmt":"2025-01-14T20:15:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6043"},"modified":"2025-01-15T05:15:00","modified_gmt":"2025-01-14T20:15:00","slug":"centurio-on-drivers-of-multilingual-ability-of-large-vision-language-model","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6043","title":{"rendered":"Centurio: On Drivers of Multilingual Ability of Large Vision-Language Model\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Centurio: On Drivers of Multilingual Ability of Large Vision-Language Model\u00a0<\/strong>[66.2]<br>\u591a\u304f\u306eLVLM(Large Vision-Language Models)\u306f\u3001\u4e3b\u306b\u82f1\u8a9e\u306e\u30c7\u30fc\u30bf\u306b\u57fa\u3065\u3044\u3066\u8a13\u7df4\u3055\u308c\u3066\u3044\u308b\u3002 \u7570\u306a\u308b\u8a00\u8a9e\u7fa4\u306b\u5bfe\u3059\u308b\u5b66\u7fd2\u304c\u3044\u304b\u306b\u7570\u306a\u308b\u304b\u3092\u691c\u8a0e\u3059\u308b\u3002 \u79c1\u305f\u3061\u306fCenturio(100\u8a00\u8a9eLVLM)\u3092\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3057\u300114\u306e\u30bf\u30b9\u30af\u306856\u306e\u8a00\u8a9e\u3092\u5bfe\u8c61\u3068\u3057\u305f\u8a55\u4fa1\u3067\u6700\u5148\u7aef\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u63d0\u4f9b\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2501.05122v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2501.05122v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 09 Jan 2025 10:26:14 GMT)<\/li>\n\n\n\n<li>Large Vision-Language Model\u306b\u304a\u3051\u308b\u591a\u8a00\u8a9e\u5316\u306e\u691c\u8a3c\u3001\u82f1\u8a9e\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u4f4e\u4e0b\u3055\u305b\u308b\u3053\u3068\u306a\u304f\u5bfe\u5fdc\u53ef\u80fd\u306a\u8a00\u8a9e\u6570\u306a\u3069\u306b\u7126\u70b9\u3092\u5f53\u3066\u3066\u3044\u308b\u3002\u300cour analysis reveals that one can (i) include as many as 100 training languages simultaneously\u3000(ii) with as little as 25-50% of non-English data, to greatly improve multilingual performance while retaining strong English performance.\u3000We further find that (iii) including non-English OCR data in pre-training and instruction-tuning is paramount for improving multilingual text-in-image understanding.\u300d\u306f\u8208\u5473\u6df1\u3044\u7d50\u679c\u3002LLaVA\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u30012.8B\u306ePhi3.5\u3001Llama 3 8B\u3067\u306e\u691c\u8a3c\u3002<\/li>\n\n\n\n<li>\u305d\u306e\u5f8c\u3001\u300cAfter benchmarking different 7-9B parameter LLMs, we find that Aya-Expanse and Qwen 2.5 give the overall best results.\u300d\u306e\u7d50\u679c\u3001Aya-Expanse\u3068Qwen 2.5\u3092\u7528\u3044\u3066\u30e2\u30c7\u30eb\u69cb\u7bc9\u3092\u884c\u3063\u3066\u3044\u308b\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/gregor-ge.github.io\/Centurio\/\">Centurio: On Drivers of Multilingual Ability of Large Vision-Language Model<\/a><\/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":[251,267],"class_list":["post-6043","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-mllm","tag-multilingual"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6043","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=6043"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6043\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6043"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6043"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6043"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}