Nvidia is increasing the layout of an open source circuit and is investing resources to develop a major model. The company hopes to use this model to drive hardware demand, but this move also means that it may compete with its own customers and partners. A number of people involved in the Nemotron project revealed that Nvidia plans to build the largest basic model in the Nemotron 4 series and target performance against the world's top open source models. There are 570 authors of papers related to Nvidia's previous flagship big model, and employees say Nemotron 4 will further expand the number of people involved in R&D. A former employee said, “Everyone wants to be involved at this stage.” Recently, Nvidia has continued to launch a variety of open source models, and this research and development is based on this. On Tuesday, Nvidia released the lightweight Nemotron 3.5 Lightning model, which focuses on efficient, high-speed running smart devices. The company also launched free model routing software to help enterprises quickly build model scheduling tools — tools that can assign different AI tasks to suitable and cost-effective models. A number of Nemotron project employees revealed that the number of parameters in the flagship Nemotron 4 reached at least trillion. The number of parameters is a parameter unit that is continuously adjusted during model learning. This scale is about twice that of Nvidia's current flagship model, the Nemotron 3 Ultra. Even after reaching the trillion-dollar parameter, the model is still smaller than America's leading open source model; however, Nvidia values model compression technology. Relying on this technology, smaller models are also expected to achieve better performance.

Nvidia is increasing the layout of an open source circuit and is investing resources to develop a major model. The company hopes to use this model to drive hardware demand, but this move also means that it may compete with its own customers and partners. A number of people involved in the Nemotron project revealed that Nvidia plans to build the largest basic model in the Nemotron 4 series and target performance against the world's top open source models. There are 570 authors of papers related to Nvidia's previous flagship big model, and employees say Nemotron 4 will further expand the number of people involved in R&D. A former employee said, “Everyone wants to be involved at this stage.” Recently, Nvidia has continued to launch a variety of open source models, and this research and development is based on this. On Tuesday, Nvidia released the lightweight Nemotron 3.5 Lightning model, which focuses on efficient, high-speed running smart devices. The company also launched free model routing software to help enterprises quickly build model scheduling tools — tools that can assign different AI tasks to suitable and cost-effective models. A number of Nemotron project employees revealed that the number of parameters in the flagship Nemotron 4 reached at least trillion. The number of parameters is a parameter unit that is continuously adjusted during model learning. This scale is about twice that of Nvidia's current flagship model, the Nemotron 3 Ultra. Even after reaching the trillion-dollar parameter, the model is still smaller than America's leading open source model; however, Nvidia values model compression technology. Relying on this technology, smaller models are also expected to achieve better performance.