[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"aihot-art-84288":3},{"itemId":4,"vertical":5,"category":6,"source":7,"score":8,"title":9,"summary":10,"analysis":11,"url":12,"coverUrl":13,"direction":13,"marketSignal":13,"publishedAt":14},"84288","ai","技巧与观点","NVIDIA Technical Blog",61,"使用智能体技能在一天内后训练NVIDIA Cosmos 3","NVIDIA展示如何使用智能体技能在一天内后训练Cosmos 3模型，提升视觉推理模型准确率。","NVIDIA 展示了如何使用智能体技能在一天内完成 Cosmos 3 视觉推理模型的后训练，将准确率提升至 90% 以上。\n· 痛点解决：传统后训练需要数天处理数据格式、容器设置、训练脚本等，而 AI 代理可自动完成这些步骤。\n· 高效流程：代理自主进行数据准备、基线评估、超参数搜索，无需人工干预。\n· 显著成果：在视频任务上，后训练后模型准确率突破 90%，且几乎无需手动调优。\n· 适用场景：特别适合需要快速适配生产视频任务的开发者。\n影响\u002F看点：该技术大幅降低了视觉模型后训练的门槛，让非专家也能高效优化模型，推动 AI 在视频领域的落地。","https:\u002F\u002Fdeveloper.nvidia.com\u002Fblog\u002Fpost-train-nvidia-cosmos-3-in-one-day-using-agent-skills\u002F",null,"2026-07-15 00:00:00"]