{"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\/ro\/how-good-is-deepseeks-janus-pro\/","title":{"rendered":"C\u00e2t de bun este DeepSeek Janus-Pro?"},"content":{"rendered":"<div style=\"margin-top: 0px; margin-bottom: 0px;\" class=\"sharethis-inline-share-buttons\" ><\/div>\n<p>\u00cen ajunul Festivalului Prim\u0103verii, a fost lansat modelul DeepSeek-R1. Cu arhitectura sa RL pur\u0103, acesta a \u00eenv\u0103\u021bat din marile inova\u021bii ale CoT \u0219i surclaseaz\u0103 <a href=\"https:\/\/openai.com\/\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">ChatGPT<\/a> \u00een matematic\u0103, cod \u0219i ra\u021bionament logic.<\/p>\n\n\n\n<p>\u00cen plus, ponderile modelelor sale open-source, costurile sc\u0103zute de formare \u0219i pre\u021burile ieftine ale API au f\u0103cut din DeepSeek un hit pe internet, determin\u00e2nd chiar sc\u0103derea pre\u021burilor ac\u021biunilor NVIDIA \u0219i ASML pentru o perioad\u0103.<\/p>\n\n\n\n<p>\u00cen timp ce explodeaz\u0103 \u00een popularitate, DeepSeek a lansat, de asemenea, o versiune actualizat\u0103 a modelului multimodal mare Janus (Janus), Janus-Pro, care mo\u0219tene\u0219te arhitectura unificat\u0103 a genera\u021biei anterioare de \u00een\u021belegere \u0219i generare multimodal\u0103 \u0219i optimizeaz\u0103 strategia de formare, scal\u00e2nd datele de formare \u0219i dimensiunea modelului, aduc\u00e2nd performan\u021be mai puternice.<\/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\">Tabla de con\u021binut<\/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=\"Toggle Tabelul de con\u021binut\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/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\/ro\/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\/ro\/how-good-is-deepseeks-janus-pro\/#Model_architecture\" >Arhitectura modelului<\/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\/ro\/how-good-is-deepseeks-janus-pro\/#Training_strategy\" >Strategia de formare<\/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\/ro\/how-good-is-deepseeks-janus-pro\/#Training_data_scaling\" >Scalarea datelor de formare<\/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\/ro\/how-good-is-deepseeks-janus-pro\/#Model_scaling\" >Scalarea modelului<\/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\/ro\/how-good-is-deepseeks-janus-pro\/#Model_evaluation\" >Evaluarea modelului<\/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> este un model de limbaj multimodal unificat (MLLM) care poate procesa simultan sarcini de \u00een\u021belegere multimodal\u0103 \u0219i sarcini de generare, adic\u0103 poate \u00een\u021belege con\u021binutul unei imagini \u0219i, de asemenea, poate genera text.<\/p>\n\n\n\n<p>Acesta decupleaz\u0103 codificatoarele vizuale pentru \u00een\u021belegerea \u0219i generarea multimodal\u0103 (de exemplu, se utilizeaz\u0103 tokenizere diferite pentru intrarea \u00een\u021belegerii imaginii \u0219i pentru intrarea \u0219i ie\u0219irea gener\u0103rii imaginii) \u0219i le proceseaz\u0103 utiliz\u00e2nd un transformator autoregresiv unificat.<\/p>\n\n\n\n<p>Fiind un model avansat de \u00een\u021belegere \u0219i generare multimodal\u0103, acesta este o versiune \u00eembun\u0103t\u0103\u021bit\u0103 a modelului Janus anterior.<\/p>\n\n\n\n<p>\u00cen mitologia roman\u0103, Janus (Janus) este un zeu gardian cu dou\u0103 fe\u021be care simbolizeaz\u0103 contradic\u021bia \u0219i tranzi\u021bia. El are dou\u0103 fe\u021be, ceea ce sugereaz\u0103, de asemenea, c\u0103 modelul Janus poate \u00een\u021belege \u0219i genera imagini, ceea ce este foarte potrivit. Deci, ce anume a actualizat PRO?<\/p>\n\n\n\n<p>Janus, ca un model mic de 1.3B, este mai mult o versiune de previzualizare dec\u00e2t o versiune oficial\u0103. Acesta exploreaz\u0103 \u00een\u021belegerea \u0219i generarea multimodal\u0103 unificat\u0103, dar are multe probleme, cum ar fi efecte instabile de generare a imaginilor, abateri mari de la instruc\u021biunile utilizatorului \u0219i detalii inadecvate.<\/p>\n\n\n\n<p>Versiunea Pro optimizeaz\u0103 strategia de formare, cre\u0219te setul de date de formare \u0219i ofer\u0103 un model mai mare (7B) din care se poate alege, oferind \u00een acela\u0219i timp un model 1B.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Model_architecture\"><\/span>Arhitectura modelului<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 \u0219i Janus<\/a> sunt identice \u00een ceea ce prive\u0219te arhitectura modelului. (Doar 1,3 miliarde! Janus unific\u0103 \u00een\u021belegerea \u0219i generarea multimodal\u0103)<\/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>Principiul de proiectare de baz\u0103 este de a decupla codificarea vizual\u0103 pentru a sprijini \u00een\u021belegerea \u0219i generarea multimodal\u0103. Janus-Pro codific\u0103 separat imaginea\/textul original de intrare, extrage caracteristici \u00eenalt-dimensionale \u0219i le proceseaz\u0103 printr-un transformator autoregresiv unificat.<\/p>\n\n\n\n<p>\u00cen\u021belegerea imaginii multimodale utilizeaz\u0103 SigLIP pentru a codifica caracteristicile imaginii (codificator albastru \u00een figura de mai sus), iar sarcina de generare utilizeaz\u0103 tokenizatorul VQ pentru a discretiza imaginea (codificator galben \u00een figura de mai sus). \u00cen cele din urm\u0103, toate secven\u021bele de caracteristici sunt introduse \u00een LLM pentru procesare<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Training_strategy\"><\/span>Strategia de formare<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>\u00cen ceea ce prive\u0219te strategia de formare, Janus-Pro a adus mai multe \u00eembun\u0103t\u0103\u021biri. Versiunea veche a Janus folosea o strategie de formare \u00een trei etape, \u00een care etapa I preg\u0103te\u0219te adaptorul de intrare \u0219i capul de generare a imaginii pentru \u00een\u021belegerea \u0219i generarea imaginii, etapa II efectueaz\u0103 o preformare unificat\u0103, iar etapa III ajusteaz\u0103 codificatorul de \u00een\u021belegere pe aceast\u0103 baz\u0103. (Strategia de formare Janus este prezentat\u0103 \u00een figura de mai jos).<\/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>Cu toate acestea, aceast\u0103 strategie utilizeaz\u0103 metoda PixArt pentru a \u00eemp\u0103r\u021bi formarea gener\u0103rii text-imagine \u00een etapa II, ceea ce duce la o eficien\u021b\u0103 de calcul sc\u0103zut\u0103.<\/p>\n\n\n\n<p>\u00cen acest scop, am prelungit timpul de formare din etapa I \u0219i am ad\u0103ugat formarea cu date ImageNet, astfel \u00eenc\u00e2t modelul s\u0103 poat\u0103 modela eficient dependen\u021bele pixelilor cu parametri LLM stabili. \u00cen etapa a II-a, am eliminat datele ImageNet \u0219i am utilizat direct datele perechii text-imagine pentru formare, ceea ce \u00eembun\u0103t\u0103\u021be\u0219te eficien\u021ba form\u0103rii. \u00cen plus, am ajustat raportul de date \u00een etapa III (date multimodale:doar text:graf vizual-semantic de la 7:3:10 la 5:1:4), \u00eembun\u0103t\u0103\u021bind \u00een\u021belegerea multimodal\u0103 \u0219i men\u021bin\u00e2nd \u00een acela\u0219i timp capacit\u0103\u021bile de generare vizual\u0103.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Training_data_scaling\"><\/span>Scalarea datelor de formare<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Janus-Pro scaleaz\u0103, de asemenea, datele de instruire ale Janus \u00een ceea ce prive\u0219te \u00een\u021belegerea multimodal\u0103 \u0219i generarea vizual\u0103.<\/p>\n\n\n\n<p>\u00cen\u021belegerea multimodal\u0103: Datele de preantrenare din etapa a II-a se bazeaz\u0103 pe DeepSeek-VL2 \u0219i includ aproximativ 90 de milioane de e\u0219antioane noi, inclusiv date privind legendele imaginilor (cum ar fi YFCC) \u0219i date privind \u00een\u021belegerea tabelelor, graficelor \u0219i documentelor (cum ar fi Docmatix).<\/p>\n\n\n\n<p>Etapa III de reglaj fin supravegheat introduce \u00een continuare \u00een\u021belegerea MEME, date de dialog chinezesc etc., pentru a \u00eembun\u0103t\u0103\u021bi performan\u021ba modelului \u00een ceea ce prive\u0219te procesarea multitask \u0219i capacit\u0103\u021bile de dialog.<\/p>\n\n\n\n<p>Generarea vizual\u0103: Versiunile anterioare foloseau date reale de calitate sc\u0103zut\u0103 \u0219i zgomot ridicat, care afectau stabilitatea \u0219i estetica imaginilor generate de text.<\/p>\n\n\n\n<p>Janus-Pro introduce aproximativ 72 de milioane de date estetice sintetice, aduc\u00e2nd raportul dintre datele reale \u0219i datele sintetice la 1:1. Experimentele au ar\u0103tat c\u0103 datele sintetice accelereaz\u0103 convergen\u021ba modelului \u0219i \u00eembun\u0103t\u0103\u021besc semnificativ stabilitatea \u0219i calitatea estetic\u0103 a imaginilor generate.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Model_scaling\"><\/span>Scalarea modelului<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Janus Pro extinde dimensiunea modelului la 7B, \u00een timp ce versiunea anterioar\u0103 a lui Janus a utilizat 1.5B DeepSeek-LLM pentru a verifica eficacitatea decupl\u0103rii codific\u0103rii vizuale. Experimentele arat\u0103 c\u0103 un LLM mai mare accelereaz\u0103 semnificativ convergen\u021ba \u00een\u021belegerii multimodale \u0219i a gener\u0103rii vizuale, verific\u00e2nd \u00een continuare scalabilitatea puternic\u0103 a metodei.<\/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>Experimentul utilizeaz\u0103 DeepSeek-LLM (1.5B \u0219i 7B, suport\u00e2nd o secven\u021b\u0103 maxim\u0103 de 4096) ca model lingvistic de baz\u0103. Pentru sarcina de \u00een\u021belegere multimodal\u0103, SigLIP-Large-Patch16-384 este utilizat ca codificator vizual, dimensiunea dic\u021bionarului codificatorului este 16384, multiplul de downsampling al imaginii este 16, iar adaptoarele de \u00een\u021belegere \u0219i de generare sunt MLP cu dou\u0103 straturi.<\/p>\n\n\n\n<p>Etapa II de formare utilizeaz\u0103 o strategie de oprire timpurie 270K, toate imaginile sunt ajustate uniform la o rezolu\u021bie de 384 \u00d7 384 \u0219i se utilizeaz\u0103 ambalarea secven\u021belor pentru a \u00eembun\u0103t\u0103\u021bi eficien\u021ba form\u0103rii . Janus-Pro este antrenat \u0219i evaluat utiliz\u00e2nd HAI-LLM. Versiunile 1.5B\/7B au fost antrenate pe 16\/32 de noduri (8\u00d7Nvidia A100 40GB per nod) timp de 9\/14 zile, respectiv.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Model_evaluation\"><\/span>Evaluarea modelului<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Janus-Pro a fost evaluat separat \u00een ceea ce prive\u0219te \u00een\u021belegerea \u0219i generarea multimodal\u0103. \u00cen general, \u00een\u021belegerea poate fi u\u0219or slab\u0103, dar este considerat\u0103 excelent\u0103 \u00een r\u00e2ndul modelelor open source de aceea\u0219i dimensiune (se presupune c\u0103 este \u00een mare m\u0103sur\u0103 limitat\u0103 de rezolu\u021bia fix\u0103 de intrare \u0219i de capacit\u0103\u021bile OCR).<\/p>\n\n\n\n<p>Janus-Pro-7B a ob\u021binut 79,2 puncte \u00een testul de referin\u021b\u0103 MMBench, care este aproape de nivelul modelelor open source de prim rang (aceea\u0219i dimensiune a InternVL2.5 \u0219i Qwen2-VL este de aproximativ 82 de puncte). Cu toate acestea, este o \u00eembun\u0103t\u0103\u021bire bun\u0103 fa\u021b\u0103 de genera\u021bia anterioar\u0103 de Janus.<\/p>\n\n\n\n<p>\u00cen ceea ce prive\u0219te generarea de imagini, \u00eembun\u0103t\u0103\u021birea fa\u021b\u0103 de genera\u021bia anterioar\u0103 este \u0219i mai semnificativ\u0103 \u0219i este considerat\u0103 a fi un nivel excelent printre modelele open source. Scorul lui Janus-Pro \u00een testul de referin\u021b\u0103 GenEval (0,80) dep\u0103\u0219e\u0219te, de asemenea, modele precum DALL-E 3 (0,67) \u0219i Stable Diffusion 3 Medium (0,74).<\/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>\u00cen ajunul Festivalului Prim\u0103verii, a fost lansat modelul DeepSeek-R1. Cu arhitectura sa RL pur\u0103, acesta a \u00eenv\u0103\u021bat din marile inova\u021bii ale CoT \u0219i dep\u0103\u0219e\u0219te ChatGPT \u00een matematic\u0103, cod \u0219i ra\u021bionament logic. \u00cen plus, greut\u0103\u021bile modelului s\u0103u open-source, costurile sc\u0103zute de formare \u0219i pre\u021burile ieftine ale API au f\u0103cut din DeepSeek un hit pe internet, chiar...<\/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\/ro\/wp-json\/wp\/v2\/posts\/847","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/janusai.pro\/ro\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/janusai.pro\/ro\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/janusai.pro\/ro\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/janusai.pro\/ro\/wp-json\/wp\/v2\/comments?post=847"}],"version-history":[{"count":1,"href":"https:\/\/janusai.pro\/ro\/wp-json\/wp\/v2\/posts\/847\/revisions"}],"predecessor-version":[{"id":856,"href":"https:\/\/janusai.pro\/ro\/wp-json\/wp\/v2\/posts\/847\/revisions\/856"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/janusai.pro\/ro\/wp-json\/wp\/v2\/media\/704"}],"wp:attachment":[{"href":"https:\/\/janusai.pro\/ro\/wp-json\/wp\/v2\/media?parent=847"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/janusai.pro\/ro\/wp-json\/wp\/v2\/categories?post=847"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/janusai.pro\/ro\/wp-json\/wp\/v2\/tags?post=847"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}