用 NVIDIA Dynamo-Triton 部署 HSTU 生成式推荐模型
原标题：Deploying an HSTU Generative Recommender with NVIDIA Dynamo-Triton
AI 摘要
NVIDIA 发布技术博客，介绍如何通过 NVIDIA Dynamo-Triton（原 Triton Inference Server）部署 HSTU 生成式推荐模型。该工作流结合 PyTorch AOTI 编译、FlexKV 支持的 KV 缓存、NV embedding cache 和原生 C++ 验证，在 RTX PRO 6000 Blackwell 上实现最高 5.93 倍延迟降低。文章展示了从 PyTorch 开发到生产推理的完整路径，无需为独立运行时重写模型。
正文节选
Generative recommender (GR) systems are emerging as a powerful new approach for large-scale personalization. Instead of treating recommendation as a set of isolated retrieval, ranking, and prediction stages, GRs reformulate recommendation as sequence modeling over user behavior. A user’s interactions, context, candidate items, and actions become tokens in a high-cardinality event stream, and the model learns to generate or score the next relevant items from that sequence. This approach is especi