{"id":110383,"date":"2025-11-12T20:12:33","date_gmt":"2025-11-12T20:12:33","guid":{"rendered":"https:\/\/ekamu.net\/?p=110383"},"modified":"2025-11-12T20:12:33","modified_gmt":"2025-11-12T20:12:33","slug":"baidu-gpt-5i-geride-birakan-yapay-zeka-modelini-tanitti","status":"publish","type":"post","link":"https:\/\/ekamu.net\/index.php\/2025\/11\/12\/baidu-gpt-5i-geride-birakan-yapay-zeka-modelini-tanitti\/","title":{"rendered":"Baidu, GPT-5’i geride b\u0131rakan yapay zeka modelini tan\u0131tt\u0131"},"content":{"rendered":"

<\/span> Baidu, ERNIE-4.5-VL-28B-A3B-Thinking<\/strong> ad\u0131n\u0131 verdi\u011fi yeni a\u00e7\u0131k kaynakl\u0131 \u00e7ok modlu yapay zeka modelini tan\u0131tt\u0131. \u00c7inli teknoloji devi, yeni modelinin OpenAI\u2019nin GPT-5’i<\/strong>\u00a0ve Google\u2019\u0131n Gemini 2.5 Pro<\/strong>‘sunu\u00a0bir\u00e7ok \u00f6l\u00e7\u00fcmde geride b\u0131rakt\u0131\u011f\u0131n\u0131 iddia ediyor.\u00a0 <\/figure>\n

Verimli yap\u0131s\u0131yla \u00f6ne \u00e7\u0131k\u0131yor<\/b><\/p>\n

Esnek Apache 2.0 lisans\u0131 alt\u0131nda yay\u0131nlanan model, toplam 28 milyar parametreye <\/strong>sahip bir “Uzmanlar Kar\u0131\u015f\u0131m\u0131” (Mixture of Experts)<\/strong> mimarisini kullan\u0131yor.\u00a0Toplamda 28 milyar parametreye sahip olmas\u0131na ra\u011fmen, her \u00e7al\u0131\u015ft\u0131rmada yaln\u0131zca 3 milyar parametre<\/strong> etkinle\u015fiyor. Bu sayede, y\u00fcksek performans\u0131 korurken hesaplama maliyetleri ciddi \u00f6l\u00e7\u00fcde azal\u0131yor.<\/strong> Baidu\u2019nun a\u00e7\u0131klamas\u0131na g\u00f6re model, Gemini 2.5 Pro\u2019yu belge ve grafik anlama testleri<\/strong> olan VQA, MMBench ve SEED-Bench k\u0131yaslamalar\u0131nda geride b\u0131rak\u0131yor.<\/p>\n

Bu hafif tasar\u0131m, ERNIE-4.5-VL\u2019nin benzer tam parametreli modellere g\u00f6re iki ila \u00fc\u00e7 kat daha h\u0131zl\u0131 \u00e7\u0131kar\u0131m<\/strong> yapmas\u0131n\u0131 sa\u011fl\u0131yor ve \u00e7ok daha az bellek kullan\u0131yor. Modelin \u00f6ne \u00e7\u0131kan yenili\u011fi olan \u201cG\u00f6rsellerle D\u00fc\u015f\u00fcnme” <\/strong>\u00f6zelli\u011fi, g\u00f6r\u00fcnt\u00fclerin belirli b\u00f6lgelerine yak\u0131nla\u015farak o k\u0131s\u0131mlar\u0131 ayr\u0131 ayr\u0131 analiz edip, bu yerel g\u00f6zlemleri genel bir sonuca entegre edebilmesini sa\u011fl\u0131yor.<\/p>\n

Baidu ara\u015ft\u0131rmac\u0131lar\u0131, modeli b\u00fcy\u00fck \u00f6l\u00e7ekli g\u00f6rsel-dilsel ak\u0131l y\u00fcr\u00fctme verileriyle uzun bir ara\u00a0e\u011fitim\u00a0s\u00fcrecinden ge\u00e7irerek e\u011fitti. E\u011fitimde GSPO ve IcePop gibi ileri d\u00fczey peki\u015ftirmeli \u00f6\u011frenme y\u00f6ntemleri ile dinamik zorluk \u00f6rneklemesi teknikleri kullan\u0131ld\u0131. Sonu\u00e7 olarak model, g\u00f6rsel ak\u0131l y\u00fcr\u00fctme, STEM problemleri \u00e7\u00f6zme, g\u00f6rsel ba\u011flama oturtma, ara\u00e7 kullan\u0131m\u0131 ve video anlama\u00a0<\/strong>konular\u0131nda \u00f6ne \u00e7\u0131k\u0131yor.<\/p>\n

\u00c7in’in a\u00e7\u0131k kaynak stratejisi<\/b><\/p>\n

Bu hamle, \u00c7in\u2019in a\u00e7\u0131k kaynakl\u0131 yapay zeka stratejisini<\/strong> h\u0131zland\u0131rd\u0131\u011f\u0131 bir d\u00f6neme denk geliyor. Son analizlere g\u00f6re \u00c7in, a\u00e7\u0131k kaynakl\u0131 AI model indirmelerinde ABD\u2019yi ge\u00e7mi\u015f durumda. DeepSeek<\/strong>, Alibaba<\/strong> ve Baidu<\/strong> gibi \u015firketler, Bat\u0131l\u0131 rakiplerine k\u0131yasla g\u00fc\u00e7l\u00fc modelleri daha h\u0131zl\u0131 ve daha d\u00fc\u015f\u00fck maliyetle<\/strong> geli\u015ftirebildiklerini g\u00f6steriyor. Bu da, \u00fccretli API\u2019lar ve kapal\u0131 modeller \u00fczerinden gelir elde eden OpenAI ve Anthropic gibi firmalar \u00fczerinde bask\u0131 olu\u015fturuyor.<\/p>\n\n

<\/span><\/p>\n

Kaynak :\u00a0https:\/\/www.donanimhaber.com\/baidu-gpt-5-i-geride-birakan-yapay-zeka-modelini-tanitti–198489<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

Baidu, ERNIE-4.5-VL-28B-A3B-Thinking ad\u0131n\u0131 verdi\u011fi yeni a\u00e7\u0131k kaynakl\u0131 \u00e7ok modlu yapay zeka modelini tan\u0131tt\u0131. \u00c7inli teknoloji devi, yeni modelinin OpenAI\u2019nin GPT-5’i\u00a0ve Google\u2019\u0131n Gemini 2.5 Pro’sunu\u00a0bir\u00e7ok \u00f6l\u00e7\u00fcmde geride b\u0131rakt\u0131\u011f\u0131n\u0131 iddia ediyor …<\/p>\n","protected":false},"author":1,"featured_media":110384,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[452],"class_list":["post-110383","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-teknoloji","tag-model"],"_links":{"self":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/110383","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/comments?post=110383"}],"version-history":[{"count":1,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/110383\/revisions"}],"predecessor-version":[{"id":110386,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/110383\/revisions\/110386"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/media\/110384"}],"wp:attachment":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/media?parent=110383"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/categories?post=110383"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/tags?post=110383"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}