Lingyiwuwu released a one-stop platform for large-scale enterprise models of Magic, announcing its full embrace of DeepSeek and providing enterprise customers with a compliant and secure one-stop DeepSeek deployment solution, covering DeepSeek model deployment, application practice and model fine-tuning tools. It can seamlessly embed DeepSeek into the enterprise business system and quickly support industry model customization and application creation.
Since the beginning of this year, DeepSeek has quickly emerged from the industry and triggered heated discussions around the world, setting a new "Intelligent Manufacturing in China" banner for the development of China's artificial intelligence industry. However, in the process of implementing DeepSeek into the industry, existing "DeepSeekInside" enterprise solutions generally have core pain points such as difficulty in deployment, application, and customization. A complete solution that can systematically solve the above pain points has not yet been formed on the market.
Kai-fu Lee, founder and CEO of Zero One Thing, said, "We have chosen to launch a one-stop platform solution for the Magic Enterprise large model based on DeepSeek, providing a 'three-step implementation' for companies that want to deploy DeepSeek, allowing them to make the transition from DeepSeek usable and useful to easy-to-use."
After the "DeepSeek shock wave" at the beginning of the year, the domestic ToB market demand is currently experiencing an unprecedented surge, and the imbalance problem in the industrial chain is also increasingly apparent - in the process of implementing from general large models to enterprise business scenarios, the lack of deployment, fine-tuning, application construction and other capabilities has become a key bottleneck. According to Sinovation Ventures, with basic model capabilities and market education in place, China's large model industry has entered the "industrial deep water zone" and is in urgent need of mature enterprise-level solutions to promote the deep integration of large models from general capabilities to vertical industries.
In May 2024, Zero-One Wanshi internally conducted an internal deduction on the diminishing marginal returns of ScalingLaw, and finally decided to abandon the training of the ultra-large model Yi-X-Large originally scheduled to have trillions of parameters, and instead train the MoE (Mixed Expert) model Yi-Lightning, which is lighter and more promising for commercial implementation.
Dr. Kaifu Li said: "What Wanzhi does is fundamentally different from the shallow services that system integrators or all-in-one machine distributors can provide. Zero One Wish has complete cutting-edge large model research and development experience, so it has the ability to deliver full-chain large model solutions in a closed loop."
It is reported that in addition to the full range of DeepSeek models, Magic Enterprise's one-stop large model platform has also carefully selected and adapted the latest mainstream domestic models such as Alibaba Qwen and Zero-One Wanyi, which can support enterprises' flexible configuration and find optimal solutions for performance and cost-effectiveness.