NVIDIA VSS Blueprint 3.3 降低视觉 AI 智能体成本
原标题：Lower the Cost of Building and Running Visual AI Agents with NVIDIA VSS Blueprint 3.3
AI 摘要
NVIDIA 发布 VSS Blueprint 3.3，通过新增 Build Vision Agent 技能和自适应高效视频采样（Adaptive EVS）降低视觉 AI 智能体的构建与运行成本。Build Vision Agent 可将告警、搜索、摘要等 VSS 工作流组合为单一应用，并基于四种已验证配置快速生成部署方案；Adaptive EVS 通过剪除帧内未变化区域的视觉 token 并批量处理事件时刻，使 60 分钟摘要的 VLM 输入 token 减少 80%，同一 GPU 上并发流增加 46%。
正文节选
Vision-language models have made it possible to build visual AI agents that understand video at production scale. The harder problem is turning that capability into a maintainable system that combines ingestion, stream processing, event detection, retrieval, summarization, and reporting. The NVIDIA Metropolis Blueprint for Video Search and Summarization (VSS) and its agent skills help developers build visual AI agents faster. VSS connects vision-language models (VLMs) such as NVIDIA Cosmos, LLMs