Unverified50% confidenceFactExact time

在RAG场景下RadixAttention可将共享上下文的KV Cache计算从O(n×请求数)降低至O(n)

1

Sources

50%

Confidence

Long-term

Relevance

7/16/2026

First Seen

Sources

GPUStack in Action: Complete Guide to Day 0 Deployment of DeepSeek-V4-Flash

Related Claims

Unverified在RAG流程返回大量候选文档片段时，可先按相关性排序再截取前3-5项传入大模型以控制输入长度65% similar Unverified文本分块策略、检索召回数量（Top-K）、重排序模型（Reranker）是影响RAG系统最终效果的三个关键变量58% similar Unverified采用RAG而非直接注入全文，通过检索最相关片段能降低上下文长度并将关键信息保持在上下文近端，缓解注意力分散和指令衰减57% similar Unverified相比RAG提交prompt有硬上限，图谱工具调用允许模型按需调用任意次数以挖掘所需信息57% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/533895

API

curl https://kongchang.com/api/v1/knowledge/claims/533895

MCP

get_claim(id=533895)