Expired50% confidenceFactExact time
There is a bug in vLLM when prefix caching and DFlash2 are both enabled, which causes the last cache unit (a mixed-layout block of approximately 16 tokens) to be directly discarded and recomputed.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
9/6/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 9/20/2026(expired)
Sources
Related Entities
Related Claims
UnverifiedvLLM、TensorRT-LLM等LLM推理服务通过KV Cache机制在单次请求内缓存中间注意力计算结果，但该缓存是请求级别的，HTTP连接关闭后即被释放70% similarUnverified2015年曝光的VENOM漏洞（CVE-2015-3456）通过向QEMU模拟的虚拟软盘控制器发送特制I/O命令触发堆溢出，实现从Guest到Host的完整VM逃逸61% similarUnverifiedOPcache caches virtual machine bytecode rather than machine code, and execution still requires interpretation and dispatch by the Zend VM.61% similarUnverifiedvLLM uses PagedAttention technology to manage KV Cache, and its memory allocation and request scheduling introduce runtime non-determinism.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/864434API
curl https://kongchang.com/api/v1/knowledge/claims/864434MCP
get_claim(id=864434)