{"id":847,"date":"2025-02-04T16:27:27","date_gmt":"2025-02-04T16:27:27","guid":{"rendered":"https:\/\/janusai.pro\/?p=847"},"modified":"2025-02-04T16:27:28","modified_gmt":"2025-02-04T16:27:28","slug":"how-good-is-deepseeks-janus-pro","status":"publish","type":"post","link":"https:\/\/janusai.pro\/tr\/how-good-is-deepseeks-janus-pro\/","title":{"rendered":"DeepSeek'in Janus-Pro'si ne kadar iyi?"},"content":{"rendered":"<div style=\"margin-top: 0px; margin-bottom: 0px;\" class=\"sharethis-inline-share-buttons\" ><\/div>\n<p>Bahar Festivali arifesinde DeepSeek-R1 modeli piyasaya s\u00fcr\u00fcld\u00fc. Saf RL mimarisi ile CoT'un b\u00fcy\u00fck yeniliklerinden dersler \u00e7\u0131karm\u0131\u015f ve daha iyi performans g\u00f6stermi\u015ftir. <a href=\"https:\/\/openai.com\/\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">ChatGPT<\/a> Matematik, kod ve mant\u0131ksal ak\u0131l y\u00fcr\u00fctme konular\u0131nda.<\/p>\n\n\n\n<p>Ayr\u0131ca, a\u00e7\u0131k kaynakl\u0131 model a\u011f\u0131rl\u0131klar\u0131, d\u00fc\u015f\u00fck e\u011fitim maliyetleri ve ucuz API fiyatlar\u0131 DeepSeek'i internet genelinde bir hit haline getirmi\u015f, hatta NVIDIA ve ASML'nin hisse senedi fiyatlar\u0131n\u0131n bir s\u00fcreli\u011fine d\u00fc\u015fmesine neden olmu\u015ftur.<\/p>\n\n\n\n<p>DeepSeek, pop\u00fclerlik patlamas\u0131 ya\u015farken, \u00f6nceki nesil multimodal anlama ve \u00fcretmenin birle\u015fik mimarisini miras alan ve e\u011fitim stratejisini optimize ederek e\u011fitim verilerini ve model boyutunu \u00f6l\u00e7eklendiren ve daha g\u00fc\u00e7l\u00fc performans getiren multimodal b\u00fcy\u00fck model Janus'un (Janus) g\u00fcncellenmi\u015f bir s\u00fcr\u00fcm\u00fc olan Janus-Pro'yi de piyasaya s\u00fcrd\u00fc.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img loading=\"lazy\" decoding=\"async\" width=\"1080\" height=\"427\" src=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/56e80359-198e-4faf-981a-54b7dfe49f02.png\" alt=\"\" class=\"wp-image-850\" srcset=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/56e80359-198e-4faf-981a-54b7dfe49f02.png 1080w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/56e80359-198e-4faf-981a-54b7dfe49f02-300x119.png 300w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/56e80359-198e-4faf-981a-54b7dfe49f02-1024x405.png 1024w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/56e80359-198e-4faf-981a-54b7dfe49f02-768x304.png 768w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/56e80359-198e-4faf-981a-54b7dfe49f02-18x7.png 18w\" sizes=\"auto, (max-width: 1080px) 100vw, 1080px\" \/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img loading=\"lazy\" decoding=\"async\" width=\"1080\" height=\"522\" src=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/af7da2cf-a17d-4ac3-95ba-42252fe1a481.png\" alt=\"\" class=\"wp-image-854\" srcset=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/af7da2cf-a17d-4ac3-95ba-42252fe1a481.png 1080w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/af7da2cf-a17d-4ac3-95ba-42252fe1a481-300x145.png 300w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/af7da2cf-a17d-4ac3-95ba-42252fe1a481-1024x495.png 1024w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/af7da2cf-a17d-4ac3-95ba-42252fe1a481-768x371.png 768w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/af7da2cf-a17d-4ac3-95ba-42252fe1a481-18x9.png 18w\" sizes=\"auto, (max-width: 1080px) 100vw, 1080px\" \/><\/figure>\n\n\n\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_82_2 counter-hierarchy ez-toc-counter ez-toc-grey ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">\u0130\u00e7indekiler<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"\u0130\u00e7erik Tablosunu De\u011fi\u015ftir\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Ge\u00e7i\u015f<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewbox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewbox=\"0 0 24 24\" version=\"1.2\" baseprofile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/janusai.pro\/tr\/how-good-is-deepseeks-janus-pro\/#Janus-Pro\" >Janus-Pro<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/janusai.pro\/tr\/how-good-is-deepseeks-janus-pro\/#Model_architecture\" >Model mimarisi<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/janusai.pro\/tr\/how-good-is-deepseeks-janus-pro\/#Training_strategy\" >E\u011fitim stratejisi<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/janusai.pro\/tr\/how-good-is-deepseeks-janus-pro\/#Training_data_scaling\" >E\u011fitim verisi \u00f6l\u00e7eklendirme<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/janusai.pro\/tr\/how-good-is-deepseeks-janus-pro\/#Model_scaling\" >Model \u00f6l\u00e7eklendirme<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/janusai.pro\/tr\/how-good-is-deepseeks-janus-pro\/#Model_evaluation\" >Model de\u011ferlendirmesi<\/a><\/li><\/ul><\/nav><\/div>\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Janus-Pro\"><\/span>Janus-Pro<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p><a href=\"https:\/\/huggingface.co\/deepseek-ai\/Janus-Pro-7B\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">Janus-Pro<\/a> \u00e7ok modlu anlama g\u00f6revlerini ve \u00fcretim g\u00f6revlerini ayn\u0131 anda i\u015fleyebilen, yani bir resmin i\u00e7eri\u011fini anlayabilen ve ayn\u0131 zamanda metin \u00fcretebilen birle\u015fik bir \u00e7ok modlu dil modelidir (MLLM).<\/p>\n\n\n\n<p>\u00c7ok modlu anlama ve \u00fcretme i\u00e7in g\u00f6rsel kodlay\u0131c\u0131lar\u0131 ay\u0131r\u0131r (yani, g\u00f6r\u00fcnt\u00fc anlama giri\u015fi ve g\u00f6r\u00fcnt\u00fc \u00fcretme giri\u015fi ve \u00e7\u0131k\u0131\u015f\u0131 i\u00e7in farkl\u0131 belirte\u00e7ler kullan\u0131l\u0131r) ve bunlar\u0131 birle\u015fik bir otoregresif d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fc kullanarak i\u015fler.<\/p>\n\n\n\n<p>Geli\u015fmi\u015f bir \u00e7ok modlu anlama ve \u00fcretme modeli olarak, \u00f6nceki Janus modelinin y\u00fckseltilmi\u015f bir versiyonudur.<\/p>\n\n\n\n<p>Roma mitolojisinde Janus (Janus) \u00e7eli\u015fkiyi ve ge\u00e7i\u015fi sembolize eden iki y\u00fczl\u00fc bir koruyucu tanr\u0131d\u0131r. \u0130ki y\u00fcz\u00fc vard\u0131r, bu da Janus modelinin g\u00f6r\u00fcnt\u00fcleri anlayabildi\u011fini ve \u00fcretebildi\u011fini g\u00f6sterir ki bu \u00e7ok uygundur. Peki PRO tam olarak neyi y\u00fckseltti?<\/p>\n\n\n\n<p>Janus, 1.3B'nin k\u00fc\u00e7\u00fck bir modeli olarak, resmi bir s\u00fcr\u00fcmden \u00e7ok bir \u00f6nizleme s\u00fcr\u00fcm\u00fc gibidir. Birle\u015ftirilmi\u015f \u00e7ok modlu anlay\u0131\u015f ve \u00fcretimi ara\u015ft\u0131r\u0131r, ancak karars\u0131z g\u00f6r\u00fcnt\u00fc olu\u015fturma efektleri, kullan\u0131c\u0131 talimatlar\u0131ndan b\u00fcy\u00fck sapmalar ve yetersiz ayr\u0131nt\u0131lar gibi bir\u00e7ok sorunu vard\u0131r.<\/p>\n\n\n\n<p>Pro s\u00fcr\u00fcm\u00fc e\u011fitim stratejisini optimize eder, e\u011fitim veri setini art\u0131r\u0131r ve 1B modeli sa\u011flarken se\u00e7im i\u00e7in daha b\u00fcy\u00fck bir model (7B) sa\u011flar.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Model_architecture\"><\/span>Model mimarisi<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p><a href=\"https:\/\/huggingface.co\/deepseek-ai\/Janus-Pro-7B\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">Jaus-Pro ve Janus<\/a> model mimarisi a\u00e7\u0131s\u0131ndan ayn\u0131d\u0131r. (Sadece 1.3B! Janus \u00e7ok modlu anlama ve \u00fcretmeyi birle\u015ftirir)<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img loading=\"lazy\" decoding=\"async\" width=\"1080\" height=\"571\" src=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/60356ab0-3c6e-4017-9eba-7ee44e0a1006.png\" alt=\"\" class=\"wp-image-851\" srcset=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/60356ab0-3c6e-4017-9eba-7ee44e0a1006.png 1080w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/60356ab0-3c6e-4017-9eba-7ee44e0a1006-300x159.png 300w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/60356ab0-3c6e-4017-9eba-7ee44e0a1006-1024x541.png 1024w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/60356ab0-3c6e-4017-9eba-7ee44e0a1006-768x406.png 768w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/60356ab0-3c6e-4017-9eba-7ee44e0a1006-18x10.png 18w\" sizes=\"auto, (max-width: 1080px) 100vw, 1080px\" \/><\/figure>\n\n\n\n<p>Temel tasar\u0131m ilkesi, \u00e7ok modlu anlama ve \u00fcretimi desteklemek i\u00e7in g\u00f6rsel kodlamay\u0131 ay\u0131rmakt\u0131r. Janus-Pro, orijinal g\u00f6r\u00fcnt\u00fc\/metin girdisini ayr\u0131 ayr\u0131 kodlar, y\u00fcksek boyutlu \u00f6zellikleri \u00e7\u0131kar\u0131r ve bunlar\u0131 birle\u015fik bir otoregresif D\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fc arac\u0131l\u0131\u011f\u0131yla i\u015fler.<\/p>\n\n\n\n<p>\u00c7ok modlu g\u00f6r\u00fcnt\u00fc anlama, g\u00f6r\u00fcnt\u00fc \u00f6zelliklerini kodlamak i\u00e7in SigLIP kullan\u0131r (yukar\u0131daki \u015fekilde mavi kodlay\u0131c\u0131) ve olu\u015fturma g\u00f6revi g\u00f6r\u00fcnt\u00fcy\u00fc ayr\u0131kla\u015ft\u0131rmak i\u00e7in VQ belirte\u00e7leyiciyi kullan\u0131r (yukar\u0131daki \u015fekilde sar\u0131 kodlay\u0131c\u0131). Son olarak, t\u00fcm \u00f6zellik dizileri i\u015flenmek \u00fczere LLM'ye girilir<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Training_strategy\"><\/span>E\u011fitim stratejisi<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>E\u011fitim stratejisi a\u00e7\u0131s\u0131ndan, Janus-Pro daha fazla iyile\u015ftirme yapm\u0131\u015ft\u0131r. Janus'un eski versiyonu, A\u015fama I'in g\u00f6r\u00fcnt\u00fc anlama ve g\u00f6r\u00fcnt\u00fc olu\u015fturma i\u00e7in giri\u015f adapt\u00f6r\u00fcn\u00fc ve g\u00f6r\u00fcnt\u00fc olu\u015fturma kafas\u0131n\u0131 e\u011fitti\u011fi, A\u015fama II'nin birle\u015fik \u00f6n e\u011fitim ger\u00e7ekle\u015ftirdi\u011fi ve A\u015fama III'\u00fcn bu temelde anlama kodlay\u0131c\u0131s\u0131na ince ayar yapt\u0131\u011f\u0131 \u00fc\u00e7 a\u015famal\u0131 bir e\u011fitim stratejisi kullan\u0131yordu. (Janus e\u011fitim stratejisi a\u015fa\u011f\u0131daki \u015fekilde g\u00f6sterilmi\u015ftir).<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img loading=\"lazy\" decoding=\"async\" width=\"1080\" height=\"381\" src=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/dbf6954f-1a18-4572-a452-ec995c8af71a.png\" alt=\"\" class=\"wp-image-849\" srcset=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/dbf6954f-1a18-4572-a452-ec995c8af71a.png 1080w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/dbf6954f-1a18-4572-a452-ec995c8af71a-300x106.png 300w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/dbf6954f-1a18-4572-a452-ec995c8af71a-1024x361.png 1024w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/dbf6954f-1a18-4572-a452-ec995c8af71a-768x271.png 768w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/dbf6954f-1a18-4572-a452-ec995c8af71a-18x6.png 18w\" sizes=\"auto, (max-width: 1080px) 100vw, 1080px\" \/><\/figure>\n\n\n\n<p>Ancak bu strateji, A\u015fama II'de metinden g\u00f6r\u00fcnt\u00fcye olu\u015fturma e\u011fitimini b\u00f6lmek i\u00e7in PixArt y\u00f6ntemini kullan\u0131r ve bu da d\u00fc\u015f\u00fck hesaplama verimlili\u011fine neden olur.<\/p>\n\n\n\n<p>Bu ama\u00e7la, A\u015fama I'in e\u011fitim s\u00fcresini uzatt\u0131k ve ImageNet verileriyle e\u011fitim ekledik, b\u00f6ylece model sabit LLM parametreleriyle piksel ba\u011f\u0131ml\u0131l\u0131klar\u0131n\u0131 etkili bir \u015fekilde modelleyebildi. A\u015fama II'de ImageNet verilerini att\u0131k ve e\u011fitim i\u00e7in do\u011frudan metin-g\u00f6r\u00fcnt\u00fc \u00e7ifti verilerini kulland\u0131k, bu da e\u011fitim verimlili\u011fini art\u0131rd\u0131. Buna ek olarak, A\u015fama III'teki veri oran\u0131n\u0131 (\u00e7ok modlu:yaln\u0131zca metin:g\u00f6rsel-anlamsal grafik verileri 7:3:10'dan 5:1:4'e) ayarlad\u0131k ve g\u00f6rsel olu\u015fturma yeteneklerini korurken \u00e7ok modlu anlay\u0131\u015f\u0131 geli\u015ftirdik.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Training_data_scaling\"><\/span>E\u011fitim verisi \u00f6l\u00e7eklendirme<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Janus-Pro ayr\u0131ca Janus'un e\u011fitim verilerini \u00e7ok modlu anlama ve g\u00f6rsel olu\u015fturma a\u00e7\u0131s\u0131ndan \u00f6l\u00e7eklendirir.<\/p>\n\n\n\n<p>\u00c7ok modlu anlama: A\u015fama II \u00f6n e\u011fitim verileri DeepSeek-VL2'ye dayanmaktad\u0131r ve g\u00f6r\u00fcnt\u00fc ba\u015fl\u0131\u011f\u0131 verileri (YFCC gibi) ve tablo, grafik ve belge anlama verileri (Docmatix gibi) dahil olmak \u00fczere yakla\u015f\u0131k 90 milyon yeni \u00f6rnek i\u00e7ermektedir.<\/p>\n\n\n\n<p>A\u015fama III denetimli ince ayar a\u015famas\u0131, modelin \u00e7oklu g\u00f6rev i\u015fleme ve diyalog yeteneklerindeki performans\u0131n\u0131 art\u0131rmak i\u00e7in MEME anlay\u0131\u015f\u0131n\u0131, \u00c7ince diyalog verilerini vb. daha fazla tan\u0131tmaktad\u0131r.<\/p>\n\n\n\n<p>G\u00f6rsel \u00fcretim: \u00d6nceki s\u00fcr\u00fcmlerde d\u00fc\u015f\u00fck kaliteli ve y\u00fcksek g\u00fcr\u00fclt\u00fcl\u00fc ger\u00e7ek veriler kullan\u0131l\u0131yordu, bu da metinle olu\u015fturulan g\u00f6r\u00fcnt\u00fclerin kararl\u0131l\u0131\u011f\u0131n\u0131 ve esteti\u011fini etkiliyordu.<\/p>\n\n\n\n<p>Janus-Pro yakla\u015f\u0131k 72 milyon sentetik estetik veri sunarak ger\u00e7ek verilerin sentetik verilere oran\u0131n\u0131 1:1'e getirmektedir. Deneyler, sentetik verilerin model yak\u0131nsamas\u0131n\u0131 h\u0131zland\u0131rd\u0131\u011f\u0131n\u0131 ve \u00fcretilen g\u00f6r\u00fcnt\u00fclerin kararl\u0131l\u0131\u011f\u0131n\u0131 ve estetik kalitesini \u00f6nemli \u00f6l\u00e7\u00fcde art\u0131rd\u0131\u011f\u0131n\u0131 g\u00f6stermi\u015ftir.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Model_scaling\"><\/span>Model \u00f6l\u00e7eklendirme<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Janus Pro model boyutunu 7B'ye geni\u015fletirken, Janus'un \u00f6nceki s\u00fcr\u00fcm\u00fc g\u00f6rsel kodlaman\u0131n ayr\u0131\u015ft\u0131r\u0131lmas\u0131n\u0131n etkinli\u011fini do\u011frulamak i\u00e7in 1,5B DeepSeek-LLM kullanm\u0131\u015ft\u0131r. Deneyler, daha b\u00fcy\u00fck bir LLM'nin multimodal anlama ve g\u00f6rsel \u00fcretimin yak\u0131nsamas\u0131n\u0131 \u00f6nemli \u00f6l\u00e7\u00fcde h\u0131zland\u0131rd\u0131\u011f\u0131n\u0131 ve y\u00f6ntemin g\u00fc\u00e7l\u00fc \u00f6l\u00e7eklenebilirli\u011fini daha da do\u011frulad\u0131\u011f\u0131n\u0131 g\u00f6stermektedir.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img loading=\"lazy\" decoding=\"async\" width=\"864\" height=\"352\" src=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/a19590e2-1805-493d-85e3-09c9b8e2274b.png\" alt=\"\" class=\"wp-image-848\" srcset=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/a19590e2-1805-493d-85e3-09c9b8e2274b.png 864w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/a19590e2-1805-493d-85e3-09c9b8e2274b-300x122.png 300w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/a19590e2-1805-493d-85e3-09c9b8e2274b-768x313.png 768w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/a19590e2-1805-493d-85e3-09c9b8e2274b-18x7.png 18w\" sizes=\"auto, (max-width: 864px) 100vw, 864px\" \/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img loading=\"lazy\" decoding=\"async\" width=\"1080\" height=\"536\" src=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/c78ed17c-6e07-43ef-bfda-ae287f597bba.png\" alt=\"\" class=\"wp-image-852\" srcset=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/c78ed17c-6e07-43ef-bfda-ae287f597bba.png 1080w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/c78ed17c-6e07-43ef-bfda-ae287f597bba-300x149.png 300w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/c78ed17c-6e07-43ef-bfda-ae287f597bba-1024x508.png 1024w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/c78ed17c-6e07-43ef-bfda-ae287f597bba-768x381.png 768w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/c78ed17c-6e07-43ef-bfda-ae287f597bba-18x9.png 18w\" sizes=\"auto, (max-width: 1080px) 100vw, 1080px\" \/><\/figure>\n\n\n\n<p>Deneyde temel dil modeli olarak DeepSeek-LLM (1.5B ve 7B, maksimum 4096 diziyi destekler) kullan\u0131lm\u0131\u015ft\u0131r. \u00c7ok modlu anlama g\u00f6revi i\u00e7in g\u00f6rsel kodlay\u0131c\u0131 olarak SigLIP-Large-Patch16-384 kullan\u0131l\u0131r, kodlay\u0131c\u0131n\u0131n s\u00f6zl\u00fck boyutu 16384, g\u00f6r\u00fcnt\u00fc alt \u00f6rnekleme katsay\u0131s\u0131 16'd\u0131r ve hem anlama hem de \u00fcretme adapt\u00f6rleri iki katmanl\u0131 MLP'lerdir.<\/p>\n\n\n\n<p>A\u015fama II e\u011fitimi 270K erken durdurma stratejisi kullan\u0131r, t\u00fcm g\u00f6r\u00fcnt\u00fcler 384\u00d7384 \u00e7\u00f6z\u00fcn\u00fcrl\u00fc\u011fe e\u015fit olarak ayarlan\u0131r ve e\u011fitim verimlili\u011fini art\u0131rmak i\u00e7in dizi paketleme kullan\u0131l\u0131r. Janus-Pro, HAI-LLM kullan\u0131larak e\u011fitilmi\u015f ve de\u011ferlendirilmi\u015ftir. 1.5B\/7B s\u00fcr\u00fcmleri s\u0131ras\u0131yla 9\/14 g\u00fcn boyunca 16\/32 d\u00fc\u011f\u00fcm (d\u00fc\u011f\u00fcm ba\u015f\u0131na 8\u00d7Nvidia A100 40GB) \u00fczerinde e\u011fitilmi\u015ftir.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Model_evaluation\"><\/span>Model de\u011ferlendirmesi<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Janus-Pro \u00e7ok modlu anlama ve \u00fcretme a\u00e7\u0131s\u0131ndan ayr\u0131 ayr\u0131 de\u011ferlendirilmi\u015ftir. Genel olarak, anlama biraz zay\u0131f olabilir, ancak ayn\u0131 boyuttaki a\u00e7\u0131k kaynak modelleri aras\u0131nda m\u00fckemmel olarak kabul edilir (b\u00fcy\u00fck \u00f6l\u00e7\u00fcde sabit giri\u015f \u00e7\u00f6z\u00fcn\u00fcrl\u00fc\u011f\u00fc ve OCR yetenekleri ile s\u0131n\u0131rl\u0131 oldu\u011funu tahmin edin).<\/p>\n\n\n\n<p>Janus-Pro-7B, MMBench benchmark testinde 79.2 puan alarak birinci kademe a\u00e7\u0131k kaynak modellerin seviyesine yak\u0131n bir puan alm\u0131\u015ft\u0131r (ayn\u0131 boyuttaki InternVL2.5 ve Qwen2-VL 82 puan civar\u0131ndad\u0131r). Bununla birlikte, \u00f6nceki nesil Janus'a g\u00f6re iyi bir geli\u015fmedir.<\/p>\n\n\n\n<p>G\u00f6r\u00fcnt\u00fc olu\u015fturma a\u00e7\u0131s\u0131ndan, \u00f6nceki nesle g\u00f6re iyile\u015fme daha da \u00f6nemlidir ve a\u00e7\u0131k kaynakl\u0131 modeller aras\u0131nda m\u00fckemmel bir seviye olarak kabul edilir. Janus-Pro'nin GenEval benchmark testindeki skoru (0.80) DALL-E 3 (0.67) ve Stable Diffusion 3 Medium (0.74) gibi modelleri de a\u015fmaktad\u0131r.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img loading=\"lazy\" decoding=\"async\" width=\"1080\" height=\"827\" src=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/47aa92e1-b474-4874-956e-db210da9d349.png\" alt=\"\" class=\"wp-image-853\" srcset=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/47aa92e1-b474-4874-956e-db210da9d349.png 1080w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/47aa92e1-b474-4874-956e-db210da9d349-300x230.png 300w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/47aa92e1-b474-4874-956e-db210da9d349-1024x784.png 1024w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/47aa92e1-b474-4874-956e-db210da9d349-768x588.png 768w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/47aa92e1-b474-4874-956e-db210da9d349-16x12.png 16w\" sizes=\"auto, (max-width: 1080px) 100vw, 1080px\" \/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img loading=\"lazy\" decoding=\"async\" width=\"1080\" height=\"744\" src=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/38de369b-7f1f-4159-83a7-5f411e816d55.png\" alt=\"\" class=\"wp-image-855\" srcset=\"https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/38de369b-7f1f-4159-83a7-5f411e816d55.png 1080w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/38de369b-7f1f-4159-83a7-5f411e816d55-300x207.png 300w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/38de369b-7f1f-4159-83a7-5f411e816d55-1024x705.png 1024w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/38de369b-7f1f-4159-83a7-5f411e816d55-768x529.png 768w, https:\/\/janusai.pro\/wp-content\/uploads\/2025\/02\/38de369b-7f1f-4159-83a7-5f411e816d55-18x12.png 18w\" sizes=\"auto, (max-width: 1080px) 100vw, 1080px\" \/><\/figure>","protected":false},"excerpt":{"rendered":"<p>Bahar Festivali arifesinde DeepSeek-R1 modeli piyasaya s\u00fcr\u00fcld\u00fc. Saf RL mimarisi ile CoT'un b\u00fcy\u00fck yeniliklerinden ders alm\u0131\u015f ve matematik, kod ve mant\u0131ksal ak\u0131l y\u00fcr\u00fctmede ChatGPT'den daha iyi performans g\u00f6stermi\u015ftir. Buna ek olarak, a\u00e7\u0131k kaynakl\u0131 model a\u011f\u0131rl\u0131klar\u0131, d\u00fc\u015f\u00fck e\u011fitim maliyetleri ve ucuz API fiyatlar\u0131 DeepSeek'i internet genelinde bir hit haline getirdi, hatta...<\/p>","protected":false},"author":2,"featured_media":704,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_kadence_starter_templates_imported_post":false,"_kad_post_transparent":"","_kad_post_title":"","_kad_post_layout":"","_kad_post_sidebar_id":"","_kad_post_content_style":"","_kad_post_vertical_padding":"","_kad_post_feature":"","_kad_post_feature_position":"","_kad_post_header":false,"_kad_post_footer":false,"footnotes":""},"categories":[1],"tags":[],"class_list":["post-847","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/janusai.pro\/tr\/wp-json\/wp\/v2\/posts\/847","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/janusai.pro\/tr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/janusai.pro\/tr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/janusai.pro\/tr\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/janusai.pro\/tr\/wp-json\/wp\/v2\/comments?post=847"}],"version-history":[{"count":1,"href":"https:\/\/janusai.pro\/tr\/wp-json\/wp\/v2\/posts\/847\/revisions"}],"predecessor-version":[{"id":856,"href":"https:\/\/janusai.pro\/tr\/wp-json\/wp\/v2\/posts\/847\/revisions\/856"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/janusai.pro\/tr\/wp-json\/wp\/v2\/media\/704"}],"wp:attachment":[{"href":"https:\/\/janusai.pro\/tr\/wp-json\/wp\/v2\/media?parent=847"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/janusai.pro\/tr\/wp-json\/wp\/v2\/categories?post=847"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/janusai.pro\/tr\/wp-json\/wp\/v2\/tags?post=847"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}