Unverified50% confidenceFactExact time
PagedAttention reduces VRAM fragmentation through block-level on-demand allocation and allows different sequences to share the same physical block, saving VRAM in scenarios such as beam search, parallel sampling, and prefix sharing.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedPagedAttention supports sharing physical blocks of the same prefix (such as system prompts) across multiple requests, implementing copy-on-write to save GPU memory.65% similarUnverifiedPagedAttention manages KV Cache by splitting it into fixed-size non-contiguous pages, allowing different requests to share the same prefix cache pages and reducing GPU memory usage.63% similarUnverifiedAt the application layer, scroll anchoring can be implemented by recording the getBoundingClientRect offset of the target element before and after content insertion and manually calling scrollBy for correction, which is suitable for scenarios such as simultaneous content injection at both ends that native behavior cannot cover.63% similarUnverified检索系统将页面切分为段落、嵌入向量化，然后将最匹配的片段拉入模型上下文61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1017343API
curl https://kongchang.com/api/v1/knowledge/claims/1017343MCP
get_claim(id=1017343)