Unverified50% confidenceCautionExact time
当上下文长度扩展至百万Token时,KV Cache本身就会占用数十GB显存
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
Unverified数千亿参数模型处理128K上下文窗口时,KV Cache可能占据数十GB显存81% similarUnverified以LLaMA-2 70B为例,处理4096 token上下文、批量大小32时,KV Cache可消耗超过80GB显存81% similarUnverified一个200K上下文窗口的请求,其KV Cache可能占据数十GB显存79% similarUnverified上下文长度从4K扩展到8K时,KV Cache占用的显存会近乎翻倍79% similarVerifiedKV Cache的显存占用随序列长度线性增长,700亿参数采用GQA的模型处理10万Token时KV Cache占用可能高达数十GB76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/562694API
curl https://kongchang.com/api/v1/knowledge/claims/562694MCP
get_claim(id=562694)