完成时间:2026年3月10日
Jimmy Lin, University of Waterloo
,更多细节参见钉钉下载
Заключенные московских изоляторов провели видеоконференцию14:53
High-Performance Sparse MoE Framework: Trinity Large Thinking is a 400B-parameter sparse Mixture-of-Experts model. It employs a 4-from-256 routing approach, triggering just 13B parameters per token during inference to deliver cutting-edge intelligence with the velocity and capacity of a substantially more compact model.