摘要：我们推出了 Nemotron 3.5 Bolt 74B-A9B 24GB，这是一款混合 Mamba-Transformer 架构的视觉-语言模型（Vision-Language Model），专门针对本地 Agent 工作负载优化了每 GB 内存下的速度与智能水平。该模型源自 Nemotron 3.5 Super VL 120B-A12B 模型家族，将 Star Elastic 压缩技术与面向受限内存部署的 2-bit 量化感知蒸馏（Quantization-Aware Distillation）相结合。该模型针对推理、Agent 工具使用以及网页与终端交互进行了专门优化。我们还训练了一个 LiLiCorr 投机草稿模型（speculative drafter），并与该模型一同发布，以达到我们设定的目标推理速度。最终生成的模型在诸多通用 Agent 能力（如 GDPVal v2）、网页浏览（如 BrowseComp）和终端操作（如 Terminal-Bench 2.1）基准测试中，精度均优于具有相似内存占用和速度的主流模型；在 RTX Spark 上可实现高达 84 TPS 的相对高速吞吐，这一速度足以媲美激活参数量少得多的模型，且完全能够满足长时程 Agent 应用的需求。

Abstract:We introduce Nemotron 3.5 Bolt 74B-A9B 24GB, a hybrid Mamba-Transformer vision-language model that optimizes speed and intelligence per GB of memory used for local agentic workloads. Derived from the Nemotron 3.5 Super VL 120B-A12B model family, Nemotron 3.5 Bolt 74B-A9B 24GB combines Star Elastic compression with 2-bit Quantization-Aware Distillation for memory-constrained deployment. The model is optimized for reasoning, agentic tool use, and web & terminal interactions. We also trained a LiLiCorr speculative drafter that is released alongside the model to reach our target inference speed. The resulting model achieves a higher accuracy than leading models with similar memory footprints and speed across many benchmarks for general agentic capabilities (e.g., GDPVal v2), web browsing (e.g., BrowseComp), and terminal use (e.g., Terminal-Bench 2.1), with a relatively high TPS of 84 on RTX Spark that is comparable to models with much lower active parameter counts and more than sufficient for long-horizon agentic applications.