Unverified50% confidenceFactExact time
PagedAttention 由 UC Berkeley vLLM 团队于 2023 年提出，可消除传统连续预分配模式下高达 60-80% 的显存碎片浪费
1
Sources
50%
Confidence
Long-term
Relevance
7/21/2026
First Seen
Sources
Related Claims
UnverifiedContinuous batching was systematically proposed in the 2023 PagedAttention paper from UC Berkeley.74% similarUnverifiedvLLM was released in 2023 by the UC Berkeley LMSYS team, and PagedAttention draws on the operating system's virtual memory paging concept to address the KV Cache VRAM waste problem.72% similarUnverifiedvLLM originated from a research team at UC Berkeley, and its core paper was published at SOSP (Symposium on Operating Systems Principles) in 2023.61% similarUnverifiedPagedAttention 是 vLLM 团队在 2023 年 SOSP 会议上发表的核心创新58% similarUnverifiedMemGPT由UC Berkeley于2023年提出，对Agent记忆架构进行系统性工程化探索58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/584868API
curl https://kongchang.com/api/v1/knowledge/claims/584868MCP
get_claim(id=584868)