{"id":3377,"date":"2024-01-30T09:51:41","date_gmt":"2024-01-30T01:51:41","guid":{"rendered":"https:\/\/www.1ai.net\/?p=3377"},"modified":"2024-01-30T09:51:41","modified_gmt":"2024-01-30T01:51:41","slug":"%e7%99%be%e5%b7%9d%e6%99%ba%e8%83%bd%e5%8f%91%e5%b8%83%e8%b6%85%e5%8d%83%e4%ba%bf%e5%8f%82%e6%95%b0%e5%a4%a7%e6%a8%a1%e5%9e%8b-baichuan-3%ef%bc%8c%e5%8f%b7%e7%a7%b0%e4%b8%ad%e6%96%87%e8%af%84%e6%b5%8b","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/3377.html","title":{"rendered":"Baichuan Intelligence releases Baichuan 3, a large model with over 100 billion parameters, claiming to surpass GPT-4 in Chinese evaluation"},"content":{"rendered":"

Baichuan Intelligence<\/a>Releasing over a hundred billion parameters ofLarge Language Model<\/a> Baichuan 3, in reviews such as CMMLU, GAOKAO and AGI-Eval.Baichuan 3 is claimed to have surpassed the Chinese task of GPT-4<\/a><\/span>.<\/p>\n

\"e5de90fb-d205-40f0-b975-7a3e1a282894\"<\/p>\n

\"2348230f-23f5-43a0-985c-3c509d18ddd8\"<\/p>\n

On medical reviews such as MCMLE, MedExam, and CMExam, which test logical reasoning skills, Baichuan 3's Chinese language results are likewise claimed to exceed the GPT-4, and are \"The best performing large model for Chinese medical tasks<\/span>\u201d.<\/p>\n

\"1a190fe7-844d-4dc8-9636-9945d0707a0f\"<\/p>\n

It is reported that Baichuan Intelligence has proposed various technical means and programs such as \"dynamic data selection\", \"importance maintenance\" and \"asynchronous CheckPoint storage\" in the training process of Baichuan 3. During the training process of Baichuan 3, Baichuan Intelligence proposed a variety of technical means and solutions, such as \"dynamic data selection\", \"importance retention\" and \"asynchronous CheckPoint storage\", with a stable training time of more than one month and a fault recovery time of less than 10 minutes.<\/p>\n

Baichuan 3 also breaks through the \"iterative reinforcement learning\" technology, which further improves the semantic understanding and generation capability, according to Baichuan's official statement.Enhancements in formatting, rhyming, and ideograms in poetry writing<\/span>For Song Lyrics, which is a difficult genre with a varied format, a deep and detailed structure, and rich rhymes, the content generated can also be neat and harmonious, so that everyone can create pentameter poems and seven-character stanzas that sing of objects and thoughts, and write \"Qinyuanchun\" and \"Dingfengbo\" that speak of aspirations and express emotions. \".<\/p>\n

\"406e4ac2-a179-445c-9087-bf4359ca1df9\"<\/p>\n

\"8411d0f5-4ba6-48dd-99b3-481e6f339eff\"<\/p>\n

Baichuan Intelligence was founded on April 10, 2023, by former Sogou CEO Wang Xiaochuan. At present, Baichuan 3 models have been online on the official website of Baichuan Intelligence, interested IT home partners can go to experience.<\/p>","protected":false},"excerpt":{"rendered":"

Baichuan Intelligence released Baichuan 3, a large language model with over 100 billion parameters, which is claimed to surpass GPT-4 in Chinese language tasks in CMMLU, GAOKAO, and AGI-Eval. Baichuan 3 is also claimed to be the \"best performing large model for Chinese language medical tasks\" in MCMLE, MedExam, and CMExam, which are used to test logical reasoning ability. In medical evaluation tests such as MCMLE, MedExam, and CMExam, Baichuan 3's Chinese language results are also claimed to exceed GPT-4, making it the \"best-performing large model for Chinese medical tasks\". According to the introduction, Baichuan Intelligence proposed various technical means and methods such as \"dynamic data selection\", \"importance maintenance\" and \"asynchronous CheckPoint storage\" in the training process of Baichuan 3. During the training process of Baichuan 3, Baichuan Intelligence proposed various technical means and solutions, such as \"dynamic data selection\", \"importance maintenance\" and \"asynchronous CheckPoint storage\", to stabilize the training time<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[146],"tags":[510,706,428],"collection":[],"class_list":["post-3377","post","type-post","status-publish","format-standard","hentry","category-news","tag-gpt-4","tag-706","tag-428"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/3377","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=3377"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/3377\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=3377"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=3377"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=3377"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=3377"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}