{"id":114737,"date":"2024-04-03T00:25:44","date_gmt":"2024-04-02T22:25:47","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/databricks-otkryl-bolshuyu-yazykovuyu-model-dbrx-operezhayushhuyu-v-testah-gpt-3-5"},"modified":"2024-04-03T00:25:44","modified_gmt":"2024-04-02T22:25:47","slug":"databricks-otkryl-bolshuyu-yazykovuyu-model-dbrx-operezhayushhuyu-v-testah-gpt-3-5","status":"publish","type":"post","link":"https:\/\/prohoster.info\/cs\/blog\/news\/databricks-otkryl-bolshuyu-yazykovuyu-model-dbrx-operezhayushhuyu-v-testah-gpt-3-5","title":{"rendered":"Databricks odhalil velk\u00fd jazykov\u00fd model DBRX, kter\u00fd v testech p\u0159ekon\u00e1v\u00e1 GPT-3.5","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Spole\u010dnost Databricks ozn\u00e1mila spu\u0161t\u011bn\u00ed velk\u00e9ho jazykov\u00e9ho modelu DBRX, kter\u00fd lze pou\u017e\u00edt k vytv\u00e1\u0159en\u00ed chatbot\u016f odpov\u00eddaj\u00edc\u00edch na ot\u00e1zky v p\u0159irozen\u00e9m jazyce, \u0159e\u0161en\u00ed navr\u017een\u00fdch matematick\u00fdch \u00faloh, generov\u00e1n\u00ed obsahu na dan\u00e9 t\u00e9ma a psan\u00ed k\u00f3du v r\u016fzn\u00fdch programovac\u00edch jazyc\u00edch. Model vyvinula spole\u010dnost Mosaic ML, kterou Databricks koupil za 1,3 miliardy dolar\u016f. K u\u010den\u00ed byl pou\u017eit klastr s 3072 GPU NVIDIA H100 Tensor Core. Pro spu\u0161t\u011bn\u00ed hotov\u00e9ho modelu se doporu\u010duje 320 GB pam\u011bti. <\/p>\n

P\u0159i tr\u00e9nov\u00e1n\u00ed modelu byla pou\u017eita architektura MoE (Mixture of Experts), kter\u00e1 umo\u017e\u0148uje z\u00edskat p\u0159esn\u011bj\u0161\u00ed odborn\u00e9 hodnocen\u00ed, a kolekce text\u016f a k\u00f3du o velikosti 12 TB. Velikost kontextu, kter\u00fd model DBRX zohled\u0148uje, \u010din\u00ed 32 tis\u00edc token\u016f (po\u010det token\u016f, kter\u00e9 model m\u016f\u017ee zpracovat a zapamatovat si p\u0159i generov\u00e1n\u00ed textu). Pro srovn\u00e1n\u00ed, velikost kontextu u model\u016f Google Gemini a OpenAI GPT-4 \u010din\u00ed 32 tis\u00edc token\u016f, Google Gemma \u2014 8 tis\u00edc, a u modelu GPT-4 Turbo \u2014 128 tis\u00edc. <\/p>\n

Model pokr\u00fdv\u00e1 132 miliard parametr\u016f a je rozd\u011blen na 16 expertn\u00edch s\u00edt\u00ed, z nich\u017e mohou b\u00fdt p\u0159i zpracov\u00e1n\u00ed po\u017eadavku pou\u017eity maxim\u00e1ln\u011b 4 (pokryt\u00ed maxim\u00e1ln\u011b 36 miliard parametr\u016f pro ka\u017ed\u00fd token). Pro srovn\u00e1n\u00ed model GPT-4 \u00fadajn\u011b zahrnuje 1,76 trilionu parametr\u016f, ned\u00e1vno p\u0159edstaven\u00fd model X\/Twitter Grok (X\/Twitter) \u2014 314 miliard, GPT-3.5 \u2014 175 miliard, YaLM (Yandex) \u2014 100 miliard, LLaMA (Meta) \u2014 65 miliard, GigaChat (Sber) \u2014 29 miliard, Gemma (Google) \u2014 7 miliard. <\/p>\n

Model a s n\u00edm souvisej\u00edc\u00ed komponenty jsou \u0161\u00ed\u0159eny pod licenc\u00ed Databricks Open Model License, kter\u00e1 umo\u017e\u0148uje pou\u017e\u00edvat, reprodukovat, kop\u00edrovat, m\u011bnit a vytv\u00e1\u0159et odvozen\u00e9 produkty, ale s ur\u010dit\u00fdmi omezen\u00edmi. Nap\u0159\u00edklad licence zakazuje vyu\u017e\u00edvat DBRX, odvozen\u00e9 modely a jak\u00e9koli v\u00fdstupy na jejich z\u00e1klad\u011b k vylep\u0161en\u00ed jin\u00fdch jazykov\u00fdch model\u016f odli\u0161n\u00fdch od DBRX. Licence tak\u00e9 zakazuje pou\u017e\u00edv\u00e1n\u00ed modelu v oblastech, kter\u00e9 poru\u0161uj\u00ed z\u00e1kony a p\u0159edpisy. Odvozen\u00e9 modely mus\u00ed b\u00fdt \u0161\u00ed\u0159eny pod stejnou licenc\u00ed. P\u0159i pou\u017e\u00edv\u00e1n\u00ed v produktech a slu\u017eb\u00e1ch, kter\u00e9 vyu\u017e\u00edv\u00e1 v\u00edce ne\u017e 700 milion\u016f u\u017eivatel\u016f m\u011bs\u00ed\u010dn\u011b, je nutn\u00e9 z\u00edskat zvl\u00e1\u0161tn\u00ed povolen\u00ed. <\/p>\n

Podle tvrzen\u00ed tv\u016frc\u016f modelu p\u0159ekon\u00e1v\u00e1 DBRX sv\u00fdmi vlastnostmi a schopnostmi modely GPT-3.5 od spole\u010dnosti OpenAI a Grok-1 od Twitteru a m\u016f\u017ee konkurovat modelu Gemini 1.0 Pro p\u0159i testov\u00e1n\u00ed porozum\u011bn\u00ed jazyku, schopnostem psan\u00ed k\u00f3du v programovac\u00edch jazyc\u00edch a \u0159e\u0161en\u00ed matematick\u00fdch \u00faloh. V n\u011bkter\u00fdch aplikac\u00edch, nap\u0159\u00edklad p\u0159i generov\u00e1n\u00ed SQL dotaz\u016f, se DBRX efektivitou p\u0159ibli\u017euje modelu GPT-4 Turbo, kter\u00fd dominuje na trhu. Krom\u011b toho se model od konkuruj\u00edc\u00edch slu\u017eeb vyzna\u010duje velmi rychlou prac\u00ed a umo\u017e\u0148uje generovat odpov\u011bdi t\u00e9m\u011b\u0159 okam\u017eit\u011b. Zejm\u00e9na m\u016f\u017ee DBRX generovat text rychlost\u00ed a\u017e 150 token\u016f za sekundu na u\u017eivatele, co\u017e je p\u0159ibli\u017en\u011b dvakr\u00e1t rychleji ne\u017e model LLaMA2-70B.

\"Databricks<\/center>
\"Databricks<\/center> <\/p>\n

D\u00e1le je mo\u017en\u00e9 zm\u00ednit publikaci technick\u00e9ho popisu otev\u0159en\u00e9ho velk\u00e9ho jazykov\u00e9ho modelu InternLM2, kter\u00fd je dostupn\u00fd pod licenc\u00ed Apache 2.0, a to ve variant\u00e1ch s 20, 7 a 1.8 miliardami parametr\u016f. Model se vyv\u00edj\u00ed ve \u0161anghajsk\u00e9 laborato\u0159i um\u011bl\u00e9 inteligence za \u00fa\u010dasti n\u011bkolika \u010d\u00ednsk\u00fdch univerzit a je pozoruhodn\u00fd s ohledem na 200K token\u016f kontextu a podporou nejen anglick\u00e9ho, ale i \u010d\u00ednsk\u00e9ho jazyka. V mnoha testech se model bl\u00ed\u017e\u00ed k GPT-4.

\"Databricks<\/center>
\"Databricks<\/center>
\"Databricks<\/center> <\/p>\n

Krom\u011b toho se uv\u00e1d\u00ed, \u017ee byly vyvinuty 84 nov\u00fdch jader pro n\u00e1soben\u00ed matic pro n\u00e1stroj llamafile, kter\u00fd vyv\u00edj\u00ed Mozilla a kter\u00fd umo\u017e\u0148uje vytv\u00e1\u0159et univerz\u00e1ln\u00ed spustiteln\u00e9 soubory pro spu\u0161t\u011bn\u00ed velk\u00fdch jazykov\u00fdch model\u016f strojov\u00e9ho u\u010den\u00ed (LLM). Zm\u011bny umo\u017enily v\u00fdrazn\u011b urychlit pr\u00e1ci model\u016f v llamafile p\u0159i prov\u00e1d\u011bn\u00ed na CPU. Nap\u0159\u00edklad prov\u00e1d\u011bn\u00ed model\u016f p\u0159i pou\u017e\u00edv\u00e1n\u00ed llamafile je nyn\u00ed rychlej\u0161\u00ed ne\u017e p\u0159i pou\u017e\u00edv\u00e1n\u00ed llama.cpp o 30 % a\u017e 500 % v z\u00e1vislosti na prost\u0159ed\u00ed, a v porovn\u00e1n\u00ed s knihovnou MKL se maticov\u00e9 operace, kter\u00e9 se vejdou do cache L2, nyn\u00ed prov\u00e1d\u011bj\u00ed dvakr\u00e1t rychleji v nov\u00e9 implementaci.
\n
Zdroj: opennet.ru<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"

\u041a\u043e\u043c\u043f\u0430\u043d\u0438\u044f Databricks \u043e\u0431\u044a\u044f\u0432\u0438\u043b\u0430 \u043e\u0431 \u043e\u0442\u043a\u0440\u044b\u0442\u0438\u0438 \u0431\u043e\u043b\u044c\u0448\u043e\u0439 \u044f\u0437\u044b\u043a\u043e\u0432\u043e\u0439 \u043c\u043e\u0434\u0435\u043b\u0438 DBRX, \u043a\u043e\u0442\u043e\u0440\u0430\u044f \u043c\u043e\u0436\u0435\u0442 \u043f\u0440\u0438\u043c\u0435\u043d\u044f\u0442\u044c\u0441\u044f \u0434\u043b\u044f \u0441\u043e\u0437\u0434\u0430\u043d\u0438\u044f \u0447\u0430\u0442\u0431\u043e\u0442\u043e\u0432, \u043e\u0442\u0432\u0435\u0447\u0430\u044e\u0449\u0438\u0445 \u043d\u0430 \u0432\u043e\u043f\u0440\u043e\u0441\u044b \u043d\u0430 \u0435\u0441\u0442\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u043c \u044f\u0437\u044b\u043a\u0435, \u0440\u0435\u0448\u0430\u044e\u0449\u0438\u0445 \u043f\u0440\u0435\u0434\u043b\u043e\u0436\u0435\u043d\u043d\u044b\u0435 \u043c\u0430\u0442\u0435\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0435 \u0437\u0430\u0434\u0430\u0447\u0438, \u0441\u043f\u043e\u0441\u043e\u0431\u043d\u044b\u0445 \u0433\u0435\u043d\u0435\u0440\u0438\u0440\u043e\u0432\u0430\u0442\u044c \u043a\u043e\u043d\u0442\u0435\u043d\u0442 \u043d\u0430 \u0437\u0430\u0434\u0430\u043d\u043d\u0443\u044e \u0442\u0435\u043c\u0443 \u0438 \u0441\u043e\u0437\u0434\u0430\u0432\u0430\u0442\u044c \u043a\u043e\u0434 \u043d\u0430 \u0440\u0430\u0437\u043b\u0438\u0447\u043d\u044b\u0445 \u044f\u0437\u044b\u043a\u0430\u0445 \u043f\u0440\u043e\u0433\u0440\u0430\u043c\u043c\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u044f. \u041c\u043e\u0434\u0435\u043b\u044c \u0440\u0430\u0437\u0440\u0430\u0431\u043e\u0442\u0430\u043d\u0430 \u043a\u043e\u043c\u043f\u0430\u043d\u0438\u0435\u0439 Mosaic ML, \u043a\u043e\u0442\u043e\u0440\u0430\u044f \u0431\u044b\u043b\u0430 \u043a\u0443\u043f\u043b\u0435\u043d\u0430 Databricks \u0437\u0430 1.3 \u043c\u043b\u0440\u0434 \u0434\u043e\u043b\u043b\u0430\u0440\u043e\u0432. \u0414\u043b\u044f \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u043b\u0441\u044f \u043a\u043b\u0430\u0441\u0442\u0435\u0440 […]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":1,"featured_media":114738,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[702],"tags":[],"class_list":["post-114737","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\n","aioseo_head_json":{"title":"\ud83e\udd47Databricks otev\u0159el velk\u00fd jazykov\u00fd model DBRX, kter\u00fd v testech p\u0159ekon\u00e1v\u00e1 GPT-3.5 | ProHoster","description":"Spole\u010dnost Databricks ozn\u00e1mila otev\u0159en\u00ed velk\u00e9ho jazykov\u00e9ho modelu DBRX, kter\u00fd lze pou\u017e\u00edt k vytv\u00e1\u0159en\u00ed chatbot\u016f, kte\u0159\u00ed odpov\u00eddaj\u00ed na ot\u00e1zky v p\u0159irozen\u00e9m jazyce a \u0159e\u0161\u00ed zadan\u00e9 matematick\u00e9 \u00fakoly.","canonical_url":"https:\/\/prohoster.info\/cs\/blog\/news\/databricks-otkryl-bolshuyu-yazykovuyu-model-dbrx-operezhayushhuyu-v-testah-gpt-3-5","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"cs_CZ","og:site_name":"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b","og:type":"article","og:title":"\ud83e\udd47Databricks \u043e\u0442\u043a\u0440\u044b\u043b \u0431\u043e\u043b\u044c\u0448\u0443\u044e \u044f\u0437\u044b\u043a\u043e\u0432\u0443\u044e \u043c\u043e\u0434\u0435\u043b\u044c DBRX, \u043e\u043f\u0435\u0440\u0435\u0436\u0430\u044e\u0449\u0443\u044e \u0432 \u0442\u0435\u0441\u0442\u0430\u0445 GPT-3.5 | ProHoster","og:description":"\u041a\u043e\u043c\u043f\u0430\u043d\u0438\u044f Databricks \u043e\u0431\u044a\u044f\u0432\u0438\u043b\u0430 \u043e\u0431 \u043e\u0442\u043a\u0440\u044b\u0442\u0438\u0438 \u0431\u043e\u043b\u044c\u0448\u043e\u0439 \u044f\u0437\u044b\u043a\u043e\u0432\u043e\u0439 \u043c\u043e\u0434\u0435\u043b\u0438 DBRX, \u043a\u043e\u0442\u043e\u0440\u0430\u044f \u043c\u043e\u0436\u0435\u0442 \u043f\u0440\u0438\u043c\u0435\u043d\u044f\u0442\u044c\u0441\u044f \u0434\u043b\u044f \u0441\u043e\u0437\u0434\u0430\u043d\u0438\u044f \u0447\u0430\u0442\u0431\u043e\u0442\u043e\u0432, \u043e\u0442\u0432\u0435\u0447\u0430\u044e\u0449\u0438\u0445 \u043d\u0430 \u0432\u043e\u043f\u0440\u043e\u0441\u044b \u043d\u0430 \u0435\u0441\u0442\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u043c \u044f\u0437\u044b\u043a\u0435, \u0440\u0435\u0448\u0430\u044e\u0449\u0438\u0445 \u043f\u0440\u0435\u0434\u043b\u043e\u0436\u0435\u043d\u043d\u044b\u0435 \u043c\u0430\u0442\u0435\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0435 \u0437\u0430\u0434\u0430\u0447\u0438.","og:url":"https:\/\/prohoster.info\/cs\/blog\/news\/databricks-otkryl-bolshuyu-yazykovuyu-model-dbrx-operezhayushhuyu-v-testah-gpt-3-5","og:image":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:secure_url":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:width":350,"og:image:height":350,"article:published_time":"2024-04-02T22:25:47+00:00","article:modified_time":"2024-04-02T22:25:47+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":{"post_id":"114737","title":null,"description":null,"keywords":null,"keyphrases":null,"primary_term":null,"canonical_url":null,"og_title":null,"og_description":null,"og_object_type":"default","og_image_type":"default","og_image_url":null,"og_image_width":null,"og_image_height":null,"og_image_custom_url":null,"og_image_custom_fields":null,"og_video":null,"og_custom_url":null,"og_article_section":null,"og_article_tags":null,"twitter_use_og":false,"twitter_card":"default","twitter_image_type":"default","twitter_image_url":null,"twitter_image_custom_url":null,"twitter_image_custom_fields":null,"twitter_title":null,"twitter_description":null,"schema":{"blockGraphs":[],"customGraphs":[],"default":{"data":{"Article":[],"Course":[],"Dataset":[],"FAQPage":[],"Movie":[],"Person":[],"Product":[],"ProductReview":[],"Car":[],"Recipe":[],"Service":[],"SoftwareApplication":[],"WebPage":[]},"graphName":"","isEnabled":true},"graphs":[]},"schema_type":"default","schema_type_options":null,"pillar_content":false,"robots_default":true,"robots_noindex":false,"robots_noarchive":false,"robots_nosnippet":false,"robots_nofollow":false,"robots_noimageindex":false,"robots_noodp":false,"robots_notranslate":false,"robots_max_snippet":null,"robots_max_videopreview":null,"robots_max_imagepreview":"large","priority":null,"frequency":null,"local_seo":null,"seo_analyzer_scan_date":"2026-01-22 21:17:19","breadcrumb_settings":null,"limit_modified_date":false,"reviewed_by":null,"ai":null,"created":"2026-01-22 21:17:19","updated":"2026-01-22 21:17:19","focus_keyword":null,"additional_keywords":null,"truseo_locale":null},"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/posts\/114737","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/comments?post=114737"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/posts\/114737\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/media\/114738"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/media?parent=114737"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/categories?post=114737"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/tags?post=114737"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}