{"id":169679,"date":"2026-01-16T12:06:12","date_gmt":"2026-01-16T12:06:12","guid":{"rendered":"https:\/\/kamucalisani.net\/?p=169679"},"modified":"2026-01-16T12:06:12","modified_gmt":"2026-01-16T12:06:12","slug":"google-yeni-ceviri-ailesi-translategemmayi-tanitti-chatgptye-jet-hizinda-yanit","status":"publish","type":"post","link":"https:\/\/kamucalisani.net\/index.php\/2026\/01\/16\/google-yeni-ceviri-ailesi-translategemmayi-tanitti-chatgptye-jet-hizinda-yanit\/","title":{"rendered":"Google, yeni \u00e7eviri ailesi TranslateGemma\u2019y\u0131 tan\u0131tt\u0131: ChatGPT\u2019ye jet h\u0131z\u0131nda yan\u0131t"},"content":{"rendered":"

<\/span> Google<\/strong>, a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 Gemma 3 modeli \u00fczerine in\u015fa edilen yeni \u00e7eviri modeli ailesi TranslateGemma<\/strong>\u2019y\u0131 duyurdu. TranslateGemma, aralar\u0131nda T\u00fcrk\u00e7e, \u0130ngilizce, \u0130spanyolca, Frans\u0131zca, \u00c7ince ve Hint\u00e7e gibi yayg\u0131n dillerin de bulundu\u011fu 55 farkl\u0131 dili <\/strong>destekliyor. Bu geni\u015f dil yelpazesi modeli hem geli\u015ftiriciler hem de \u00e7ok dilli uygulamalar i\u00e7in dikkat \u00e7ekici hale getiriyor. <\/figure>\n

Google\u2019\u0131n duyurusunun zamanlamas\u0131 ise dikkat \u00e7ekici. Zira bir g\u00fcn \u00f6nce OpenAI, sessiz sedas\u0131z ChatGPT Translate<\/strong> \u00f6zelli\u011fini kullan\u0131ma alm\u0131\u015ft\u0131. ChatGPT Translate, kelime kelime \u00e7eviriden ziyade ba\u011flam ve ton do\u011frulu\u011funa odaklanan bir yakla\u015f\u0131m sunuyor. Ancak yapt\u0131\u011f\u0131m denemelerde ciddi eksiklikleri oldu\u011funu tespit ettim. Zira bir \u00e7eviri arac\u0131ndan \u00e7ok \u00e7eviri g\u00f6r\u00fcn\u00fcme sokulmu\u015f sohbet robotu gibi davran\u0131yor. A\u015fa\u011f\u0131daki haberde bahsetti\u011fim ciddi sorunu g\u00f6rebilirsiniz.<\/p>\n

3 farkl\u0131 s\u00fcr\u00fcme sahip<\/b><\/p>\n

\u00d6te yandan TranslateGemma ise daha \u00e7ok model mimarisi ve performans verimlili\u011fi \u00fczerinden konumlan\u0131yor. TranslateGemma, 4B, 12B ve 27B parametre<\/strong> olmak \u00fczere \u00fc\u00e7 farkl\u0131 s\u00fcr\u00fcmle sunuluyor. Google\u2019\u0131n payla\u015ft\u0131\u011f\u0131 teknik verilere g\u00f6re 12B parametreli model, WMT24++ benchmark testlerinde, temel al\u0131nan Gemma 3 27B modelinden daha y\u00fcksek performans sergiliyor.<\/p>\n

<\/span> Google, 4B modelin mobil cihazlarda \u00e7\u0131kar\u0131m (inference) i\u00e7in optimize edildi\u011fini, 12B modelin ise t\u00fcketici s\u0131n\u0131f\u0131 diz\u00fcst\u00fc bilgisayarlar i\u00e7in uygun oldu\u011funu belirtiyor. En b\u00fcy\u00fck model olan 27B s\u00fcr\u00fcm\u00fc ise daha y\u00fcksek donan\u0131m gereksinimlerine sahip ve bulut ortam\u0131nda tek bir NVIDIA H100 gibi g\u00fc\u00e7l\u00fc bir GPU ile \u00e7al\u0131\u015ft\u0131r\u0131lmas\u0131 \u00f6neriliyor. <\/figure>\n

TranslateGemma\u2019n\u0131n yetenekleri yaln\u0131zca metinle s\u0131n\u0131rl\u0131 de\u011fil. Vistra g\u00f6r\u00fcnt\u00fc \u00e7eviri benchmark testlerinde yap\u0131lan de\u011ferlendirmeler, modelin g\u00f6r\u00fcnt\u00fcler i\u00e7indeki metinleri \u00e7evirme konusunda da ba\u015far\u0131l\u0131 oldu\u011funu ortaya koyuyor. Google, bunun i\u00e7in ince ayarlar yapmad\u0131\u011f\u0131n\u0131 s\u00f6yl\u00fcyor.<\/p>\n

E\u011fitimde Gemini detay\u0131<\/b><\/p>\n

<\/span> Bununla birlikte \u015firket, TranslateGemma\u2019daki bu y\u00fcksek \u201czeka yo\u011funlu\u011funun\u201d arkas\u0131nda iki a\u015famal\u0131 \u00f6zel bir e\u011fitim s\u00fcreci bulundu\u011funu a\u00e7\u0131klad\u0131. \u0130lk a\u015famada, Denetimli \u0130nce Ayar (Supervised Fine-Tuning) y\u00f6ntemi kullan\u0131ld\u0131. Bu s\u00fcre\u00e7te Gemma 3 modelleri, hem insan \u00e7evirmenler taraf\u0131ndan haz\u0131rlanm\u0131\u015f metinlerle hem de Gemini modelleri taraf\u0131ndan \u00fcretilmi\u015f y\u00fcksek kaliteli sentetik verilerle e\u011fitildi. <\/figure>\n

\u0130kinci a\u015famada ise Peki\u015ftirmeli \u00d6\u011frenme (Reinforcement Learning) devreye al\u0131nd\u0131. Bu a\u015famada MetricX-QE ve AutoMQM gibi geli\u015fmi\u015f \u00f6l\u00e7\u00fcm sistemlerini i\u00e7eren birden fazla \u00f6d\u00fcl modeli, \u00e7evirilerin daha do\u011fal, ak\u0131c\u0131 ve ba\u011flama uygun hale gelmesi i\u00e7in kullan\u0131ld\u0131.<\/p>\n

Nas\u0131l kullan\u0131l\u0131r?<\/b><\/p>\n

TranslateGemma modelleri, denemek veya kendi projelerinde kullanmak isteyenler i\u00e7in Kaggle ve Hugging Face platformlar\u0131 \u00fczerinden indirilebilir durumda. Ayr\u0131ca Gemma Cookbook arac\u0131l\u0131\u011f\u0131yla ke\u015ffedilebiliyor ve Vertex AI \u00fczerinden da\u011f\u0131t\u0131ma al\u0131nabiliyor.<\/p>\n\n

Kaynak\u00a0 :\u00a0https:\/\/www.donanimhaber.com\/google-yeni-ceviri-ailesi-translategemma-yi-tanitti–200940<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

Google, a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 Gemma 3 modeli \u00fczerine in\u015fa edilen yeni \u00e7eviri modeli ailesi TranslateGemma\u2019y\u0131 duyurdu. TranslateGemma, aralar\u0131nda T\u00fcrk\u00e7e, \u0130ngilizce, \u0130spanyolca, Frans\u0131zca, \u00c7ince ve Hint\u00e7e gibi yayg\u0131n dillerin de bulundu\u011fu 55 farkl\u0131 dili …<\/p>\n","protected":false},"author":1,"featured_media":169680,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[2865,449,559,11119],"class_list":["post-169679","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-teknoloji","tag-ceviri","tag-model","tag-modeli","tag-translate"],"_links":{"self":[{"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/posts\/169679","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=169679"}],"version-history":[{"count":1,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/posts\/169679\/revisions"}],"predecessor-version":[{"id":169684,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/posts\/169679\/revisions\/169684"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/media\/169680"}],"wp:attachment":[{"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/media?parent=169679"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/categories?post=169679"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/tags?post=169679"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}