Unverified50% confidenceFactExact time
The number of slot tokens in CondenseVLA is a preset hyperparameter that does not vary with the length of the history window, so decoder computation remains constant.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCondenseVLA's compression occurs at the decoding stage rather than the encoding stage, so the VLM backbone does not need to recompute for every frame.65% similarUnverifiedA cache hit requires the prefix to be completely identical and to exceed a certain length threshold (typically 1,024 tokens).63% similarUnverifiedKV Cache（键值缓存）用于避免自回归生成时对历史 token 的重复计算，但长上下文场景下预填充阶段需占用大量 HBM 显存带宽，拉长首 token 延迟（TTFT）61% similarUnverifiedPrefix caching can save 50%–80% of prefill-stage computation by reusing already-computed KV Cache, and vLLM implements fine-grained cache hit matching through a radix tree structure.60% similarUnverified上下文越长，模型需要缓存的 KV Cache 越大，KV Cache 使模型生成新 Token 时无需重新计算所有历史 Token 的注意力权重59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1073790API
curl https://kongchang.com/api/v1/knowledge/claims/1073790MCP
get_claim(id=1073790)