[
KongchangAI
]
Feed
Entities
Podcasts
Deep Dives
Developers
About
Get CLI
中文
Concept
Prefill-Decode disaggregation / 预填充解码分离
PD分离
LLM推理系统中将预填充(Prefill)阶段与解码(Decode)阶段分离部署的架构策略,通过独立扩展两个阶段的算力资源来优化整体推理吞吐量和延迟
Source Articles
SGLang v0.5.17深度解析:Kimi K3零日支持、Rust前端与5倍加载加速
Sep 11