英伟达开发万亿参数开源AI模型，挑战顶尖系统

Nvidia is reportedly developing a new family of Nemotron models with as many as one trillion parameters, placing the chipmaker in direct competition with the companies producing the most capable artificial intelligence systems. Nemotron 4 could be ready by late fall, with openly available models expected to form a central part of the release. The project expands Nvidia’s role beyond supplying the processors on which AI is built. The company already provides networking hardware, software libraries and complete data-center systems. A frontier-scale model would add another layer, giving customers a ready-made intelligence platform designed to operate efficiently across Nvidia hardware. The timing reflects pressure from rapidly improving Chinese models, many of which can be downloaded, modified and deployed privately. …