知识卡片
资源边界与灰度放量
在线服务要求实时返回,延迟预算可能只有几百毫秒(实时翻译)甚至更短(高频交易),而精度优化常与延迟、成本冲突,必须贴近生产环境实测——生产硬件与网络瓶颈和开发环境不同。离线批处理要评估整体算力成本,模型在 RAM 中的占用也要提前核算,这些检查可自动化,也值得人工权衡。即便全部通过也不该一键全量切换:先接一小股流量,确认服务表现符合预期再逐步加量,这种金丝雀放量把最后一步验证与生产监控衔接起来。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
原书出处 · 1 处
- Evaluating Model Validity and Quality · 查看原书依据
Is the computational performance of the model within allowable bounds? In systems that use online serving, models must return results on the fly, which typically means tight latency constraints must be met. For example, a model intended to do real-time language translation might have a budget of only a few hundred milliseconds to respond. A model used within the context of high-frequency stock trading might have significantly
参考来源
- 位置:《Reliable Machine Learning》第05章《Evaluating Model Validity and Quality》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch05.xhtml)
- 结论依据:原文明确阐述该主题,并给出定义、实践或反例。
- 原始内容:Is the computational performance of the model within allowable bounds? In systems that use online serving, models must return results on the fly, which typically means tight latency constraints must be met. For example, a model intended to do real-time language translation might have a budget of only a few hundred milliseconds to respond. A model used within the context of high-frequency stock trading might have significantly