{"id":125359,"date":"2025-10-14T13:00:16","date_gmt":"2025-10-14T13:00:16","guid":{"rendered":"https:\/\/kamucalisani.net\/?p=125359"},"modified":"2025-10-14T13:00:16","modified_gmt":"2025-10-14T13:00:16","slug":"nostaljik-donanimda-modern-teknoloji-pentium-iide-yapay-zeka-calistirildi","status":"publish","type":"post","link":"https:\/\/kamucalisani.net\/index.php\/2025\/10\/14\/nostaljik-donanimda-modern-teknoloji-pentium-iide-yapay-zeka-calistirildi\/","title":{"rendered":"Nostaljik donan\u0131mda modern teknoloji: Pentium II’de yapay zeka \u00e7al\u0131\u015ft\u0131r\u0131ld\u0131"},"content":{"rendered":"

<\/span> Oxford \u00dcniversitesi’nden ara\u015ft\u0131rmac\u0131 ve m\u00fchendislerden olu\u015fan EXO Labs<\/strong> ekibi, 1998 <\/strong>y\u0131l\u0131nda piyasaya \u00e7\u0131kan eski bir Windows 98 <\/strong>bilgisayarda g\u00fc\u00e7l\u00fc bir dil modelini \u00e7al\u0131\u015ft\u0131rmay\u0131 ba\u015fard\u0131. Hem de sadece 128MB RAM<\/strong>‘e sahip, 350MHz h\u0131z\u0131nda bir Pentium II <\/strong>i\u015flemci ile! <\/figure>\n

EXO Labs, bu deneyimi bir video ile sosyal medya platformu X’te payla\u015ft\u0131. Videoda, tozlanm\u0131\u015f bir Elonex Pentium II sistemine y\u00fcklenen Llama2.c<\/strong> tabanl\u0131 bir modelin, bir hikaye olu\u015fturma komutunu ba\u015far\u0131yla yerine getirdi\u011fi g\u00f6r\u00fcld\u00fc. \u00dcstelik h\u0131z a\u00e7\u0131s\u0131ndan da tatmin edici bir performans sergiledi.<\/p>\n

Ba\u015far\u0131 b\u00fcy\u00fck, ama\u00e7 farkl\u0131<\/b><\/p>\n

Ekibin kar\u015f\u0131la\u015ft\u0131\u011f\u0131 en b\u00fcy\u00fck zorluklardan biri, 1998 y\u0131l\u0131na ait bir i\u015fletim sistemi \u00fczerinde modern bir yapay zeka modelini derleyip \u00e7al\u0131\u015ft\u0131rmakt\u0131. Ancak bu zorluklar\u0131n \u00fcstesinden gelerek 260.000 parametreli <\/strong>bir LLaMA modelini saniyede 39.31 token <\/strong>i\u015fleme h\u0131z\u0131nda \u00e7al\u0131\u015ft\u0131rmay\u0131 ba\u015fard\u0131lar. Daha b\u00fcy\u00fck modellerde ise performans kayb\u0131 ya\u015fand\u0131; \u00f6rne\u011fin, 1 milyar parametreli bir model saniyede sadece 0.0093 token h\u0131z\u0131na ula\u015fabildi.<\/p>\n

<\/span> Peki, bu deneyin amac\u0131 neydi? EXO Labs, yapay zeka modellerini en m\u00fctevaz\u0131 donan\u0131mlarda bile \u00e7al\u0131\u015ft\u0131rabilecek hale getirerek, bu teknolojiyi evrenselle\u015ftirmeyi hedefliyor.<\/strong> G\u00fcn\u00fcm\u00fczde yapay zeka teknolojisi, genellikle g\u00fc\u00e7l\u00fc sunucu altyap\u0131lar\u0131na ve pahal\u0131 GPU’lara ba\u011f\u0131ml\u0131. Ancak EXO Labs, bu durumu tersine \u00e7evirmek i\u00e7in \u00e7al\u0131\u015f\u0131yor. <\/figure>\n

EXO Labs, “BitNet<\/strong>” ad\u0131n\u0131 verdikleri bir transformer mimarisi \u00fczerinde \u00e7al\u0131\u015f\u0131yor. Bu teknoloji, model boyutunu k\u00fc\u00e7\u00fcltmek i\u00e7in \u00fc\u00e7l\u00fc a\u011f\u0131rl\u0131klar (ternary weights) kullan\u0131yor ve 7 milyar parametreli bir modeli sadece 1.38GB depolama alan\u0131nda<\/strong> \u00e7al\u0131\u015ft\u0131rmay\u0131 m\u00fcmk\u00fcn k\u0131l\u0131yor. Daha da etkileyici olan ise, BitNet’in yaln\u0131zca CPU ile \u00e7al\u0131\u015fmak \u00fczere tasarlanm\u0131\u015f olmas\u0131. Bu mimari, 100 milyar parametreli bir modeli tek bir CPU \u00fczerinde, saniyede 5-7 token i\u015fleme h\u0131z\u0131nda \u00e7al\u0131\u015ft\u0131rabiliyor.<\/p>\n\n

Kaynak\u00a0 :\u00a0https:\/\/www.donanimhaber.com\/nostaljik-donanimda-modern-teknoloji-pentium-ii-de-yapay-zeka–185987<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

Oxford \u00dcniversitesi’nden ara\u015ft\u0131rmac\u0131 ve m\u00fchendislerden olu\u015fan EXO Labs ekibi, 1998 y\u0131l\u0131nda piyasaya \u00e7\u0131kan eski bir Windows 98 bilgisayarda g\u00fc\u00e7l\u00fc bir dil modelini \u00e7al\u0131\u015ft\u0131rmay\u0131 ba\u015fard\u0131. Hem de sadece 128MB RAM’e sahip, 350MHz h\u0131z\u0131nda bir Pentium II …<\/p>\n","protected":false},"author":1,"featured_media":125360,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[738,449],"class_list":["post-125359","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-teknoloji","tag-hiz","tag-model"],"_links":{"self":[{"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/posts\/125359","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/comments?post=125359"}],"version-history":[{"count":1,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/posts\/125359\/revisions"}],"predecessor-version":[{"id":125363,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/posts\/125359\/revisions\/125363"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/media\/125360"}],"wp:attachment":[{"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/media?parent=125359"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/categories?post=125359"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/tags?post=125359"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}