nemotron-3

【设计与创意】Build efficient open-source LLMs via hybrid Mamba-Transformer MoE architecture with LatentMoE expert design, multi-token prediction training, FP4 precision, and multi-environment RL post-training—achieving 3.3× higher throughput than equivalently-sized models while maintaining state-of-the-art reasoning, coding, and tool-use capabilities.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准