知识卡片
推理部署由资源与质量决定
交互式推理同时受模型质量、内存、计算与延迟要求约束。较小模型可降低资源和装载成本,却须用实际任务衡量效果;较大模型并不总值得额外开销。是否可在终端运行依赖模型、量化、硬件与目标时延,不能沿用书中“LLM通常不能在移动设备运行”的时期概括作永久限制。
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
- Chapter 15. AI Inference and Serving · 查看原书依据
the first thing to consider is the computational requirements of the model and your application. If your application is going to be used interactively by a user, the latency of the responses is critical to the quality of the application. Generally, users begin to notice latency at 250
参考来源
- 位置:《Designing Distributed Systems, 2nd Edition》第15章《Chapter 15. AI Inference and Serving》(来源项目 book-sources;源文件:extracted/Designing Distributed Systems, 2nd Edition/_epub-src/OEBPS/ch15.html)
- 结论依据:原文从交互性能推导硬件与小模型取舍;只保留可迁移决策维度。
- 原始内容:the first thing to consider is the computational requirements of the model and your application. If your application is going to be used interactively by a user, the latency of the responses is critical to the quality of the application. Generally, users begin to notice latency at 250