vLLM
软件
推理
智能体
The open-source serving engine that made GPU inference efficient.
您正在查看一条软件记录：AI 技术栈中的一款工具或平台——谁维护它、它运行在什么硬件上。
vLLM's PagedAttention treats the KV cache like virtual memory, ending the fragmentation that wasted most inference GPU memory and multiplying throughput. Born at UC Berkeley in 2023, it became the default open serving engine across clouds and enterprises, with hardware vendors (NVIDIA, AMD, AWS, Google) all contributing backends.
它如何与整个技术栈相连
本记录的关联链接一览——点击分组可直接跳转。
来源
记录审阅截至 2026-01-15 — 查看 方法论.