Our reporter Yuan Chuanxi
On December 22nd, the results of the first official "compliance evaluation of large model standards" in China were announced. Alibaba Cloud Tongyi Qianwen has become one of the first four domestically produced large models to pass the evaluation, meeting the relevant national standards in terms of universality, intelligence, and other dimensions. The "Compliance Evaluation of Large Model Standards" was initiated by the China Institute of Electronic Technology Standardization, aiming to establish a compliance directory for China's large model standards and lead the healthy and orderly development of the artificial intelligence industry. This evaluation has solicited opinions from dozens of top institutions in academia and industry, covering 38 specific evaluation dimensions for evaluating the universality and intelligence of language big models. It is an authoritative evaluation based on official big model testing benchmarks. Tongyi Qianwen is a large-scale language model launched by Alibaba Cloud, with features including multiple rounds of dialogue, copywriting, logical reasoning, multimodal understanding, and multilingual support. Capable of engaging in multiple rounds of interaction with humans, incorporating multimodal knowledge understanding, and possessing the ability to create copywriting. Capable of continuing to write novels, writing emails, and more. On September 13, 2023, Alibaba Cloud announced that the first batch of the Tongyi Qianwen model had been approved for filing and officially opened to the public. The Tongyi Qianwen APP has been officially launched in major mobile application markets, and everyone can directly experience the latest model capabilities through the APP. On December 1st, it was announced that Alibaba Cloud has opened up a 72 billion parameter model for Tongyi Qianwen. It is reported that among the first batch of large models that passed the evaluation, Tongyi Qianwen is the only open-source model with a wide range of developer users and enterprise customers worldwide. Its performance and security have been widely publicly tested.