Shulou(Shulou.com)11/24 Report--

Thanks to CTOnews.com netizens for the delivery of clues about the past. CTOnews.com May 27 news, today Zhihu United face wall intelligence in the "2023 China International big data Industrial Expo" released the latest results of the big language model. Li Dahai, partner and chief technology officer of Zhihu, said that Zhihu model application "search aggregation" started internal testing today, the Chinese pedestal large model CPM-Bee10b developed by facing wall intelligence was fully open source, and released the dialogue model product "face wall exposure card".

CTOnews.com previously reported that at the 2023 Zhihu Discovery Conference in April, Zhihu released the large language model "Zhihai Map AI" and internally tested the "hot list summary" of the application function of the large model on the site. A month later, Zhihu brought another large model application function "search aggregation" on the site. This product applies the large model ability to Zhihu search. Whenever the user triggers the search, the system will aggregate ideas from a large number of questions and answers to improve the efficiency of users to obtain information and form decisions. Li Dahai said that "search aggregation" started internal testing today.

("search aggregation" presentation) at the press conference, Wall-facing Intelligence joined with the OpenBMB community to open source the self-developed CPM-Bee 10b model. Li Dahai said that the model is trained independently from scratch, is based on the Transformer framework, has excellent performance in both Chinese and English, and has tens of billions of parameters and trillions of high-quality corpus. In the ZeroCLUE evaluation, CPM-Bee10b topped the list with a total score of 78.18; in the English common sense reasoning list, CPM-Bee10b scored an average score of 67, comparable to the English open source model LLaMA. "CPM-Bee10b will be fully open source and allow commercial use." Li Dahai said that the face wall intelligence has always adhered to the open source line, and will continue to embrace open source in the future, promoting the technological and ecological prosperity of the large model field.

The conference also brought the dialogue model product "Luca" developed by face-to-face intelligence, which has further performance improvement on the open source basic model, which can conduct intelligent interaction and support multiple rounds of conversations. In the live demonstration of the press conference, Luca demonstrated a number of abilities, which can not only help people understand world knowledge, deal with mathematical logic, write program code, and inspire creative inspiration, but also make use of vast amounts of knowledge and data. help people better access to information, make plans, and solve problems. The press conference also showed the multi-modal understanding ability of "Luca", which can not only analyze the landscape, geography and other picture information, but also understand the emotional meaning conveyed by character pictures. In addition, Luca can find papers and generate abstracts. According to Li Dahai, "Luca" has now started internal testing.

(face-to-face exposed card dialogue display picture)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.