ByteDance’s Doubao Large Language Model Makes Strides, Expanding into Retailand Automotive Sectors
Shanghai, China – August 21, 2024 – ByteDance’s Volcano Engine, the company’s cloud computing platform, hosted its 2024 AI Innovation Tour in Shanghai, showcasing significant upgrades to its Doubao large language model (LLM) and announcing the formation of new industry-specific AI ecosystems.
The Doubao LLM, first launched in May, has seen a remarkable surge in usage, with daily token usage exceeding 500 billion and average enterprise customer usage increasing 22-fold. This growth has fueled further development, leading to a20.3% overall improvement in the model’s capabilities compared to its initial release.
Large usage is key to refining the model, and a good model attracts more users, said Tan Dai, President of Volcano Engine. We are excited to see more AI-native and AI-transforming companies thrive on the Doubao LLM.
The latest version of Doubao boasts significant enhancements across various domains. Its role-playing ability has increased by 38.3%, enabling more natural and human-like dialogue with improved contextual awareness.Language comprehension has improved by 33.3%, making Doubao more adept at tasks like information classification, extraction, summarization, and question answering. The model has also seen advancements in long-text processing, mathematics, specialized knowledge, and code generation.
Doubao’s specialized models have also undergone significant upgrades. The Doubao text-to-image model now exhibits more precise image-text matching for long texts and excels at generating images with complex scenes, multiple subjects, intricate hand structures, and a deeper understanding of Chinese cultural elements, resulting in visually appealing Chinese-style images.
The Doubao speech recognition model, leveragingthe vast knowledge and reasoning capabilities of the LLM, has improved accuracy through contextual awareness. In multiple public tests, it outperformed other publicly released Chinese speech recognition models, achieving a maximum error reduction of 40%. The model now supports both Mandarin and various dialects, including Cantonese, Shanghainese, Sichuanese,Xianese, and Minnan.
The Doubao speech synthesis model has been upgraded with streaming speech synthesis capabilities, enabling real-time response, precise punctuation, and think-as-you-speak functionality.
Furthermore, Volcano Engine launched a real-time interactive solution for conversational AI, integrating DoubaoLLM with real-time audio and video (RTC) technology. This end-to-end solution allows businesses to seamlessly embed real-time voice capabilities into their AI applications. Users can interact with AI using voice, interrupt or interject during conversations, and experience more natural, realistic, and fluid dialogue thanks to theenhanced expressiveness and emotional nuances of the AI voice.
To cater to the high concurrency demands of enterprise production environments, the Doubao general-purpose model Pro offers a leading 800k initial TPM (tokens per minute) processing capacity, ensuring both affordability and reliability for businesses. Volcano Engine provides support for optimizingprompts and guaranteeing high concurrency, enabling clients to handle peak usage scenarios.
Beyond model advancements, Volcano Engine recognizes the importance of fostering robust ecosystems to facilitate the adoption of large language models. To this end, the company has partnered with Duoduo DMALL to establish the Retail Large Model Ecosystem Alliance. This alliance aims toempower retail businesses with low-risk access to Doubao LLM and AI capabilities, driving intelligent upgrades across the retail industry. By enhancing efficiency and innovation, the alliance aims to adapt to evolving market demands and consumer behaviors, accelerating the pace of innovation in the retail sector.
Volcano Engine is committed to collaborating withindustry partners to build a thriving retail large model ecosystem, exploring AI-driven transformations in various scenarios and extending large model applications across the retail supply chain, said Tan Dai. This will accelerate operational and turnover efficiency in retail, ultimately enhancing the consumer shopping experience.
The inaugural members of the Retail Large Model Ecosystem Alliance includeprominent players such as Wumart Group, Douyin E-commerce, Douyin Life Services, Yum! Brands, McDonald’s, China Feihe, Haidilao, Juxin Home, South 7-Eleven, Chongqing Department Store, Pagoda, Bosideng, Tianhong, Suntory, Juewei, MINISO, NielsenIQ, and Dentsu.
Dr. Zhang WenZhong, Founder of Duoduo DMALL and Wumart Group, emphasized the significance of the alliance for retail businesses, stating, The Retail Large Model Ecosystem Alliance offers a platform for collective warmth, enabling retailers toshare technical achievements and best practices, reducing costs, and making it the ideal choice for embracing AI.
We must fully embrace AI, not just for a better future but for survival, Dr. Zhang added.
In addition to the retail ecosystem, the Automotive Large Model Ecosystem Alliance has also expanded, welcoming new membersincluding Lynk & Co, Geely Galaxy, Geometry Auto, SAIC Roewe, SAIC MG, Lionheart Technology, and Da Sheng Technology.
The ongoing development of Doubao LLM and the establishment of these industry-specific ecosystems underscore Volcano Engine’s commitment to democratizing AI and empowering businesses across varioussectors to harness the transformative power of large language models. As Doubao continues to evolve and expand its reach, it is poised to play a pivotal role in shaping the future of AI-driven innovation.
【来源】https://mp.weixin.qq.com/s/nzNkPQqSTSA07OVytSOs7w
Views: 2