Unverified50% confidenceFactExact time
In loop transformers, because the weight-sharing block is applied R times, the total KV cache volume is roughly R times that of a standard model.
1
Sources
50%
Confidence
Long-term
Relevance
11/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn a standard Transformer, each layer computes KV only once; whereas in a recurrent architecture the same set of layers is executed multiple times, so looping 4 times is equivalent to storing 4 copies of the KV cache.76% similarUnverified在 Transformer 推理中,KV 缓存占用的显存可达总显存的 30%-60%73% similarUnverifiedThe KV cache stores key-value pairs from Transformer attention computations and grows linearly with sequence length.69% similarUnverifiedTransformer's KV cache size grows linearly with sequence length, whereas Mamba2's recurrent state cache size is independent of context length and depends only on the state dimension and the number of layers.66% similarUnverifiedIn on-device LLM inference, model weights, KV cache, and runtime must all fit simultaneously within the device's fixed memory.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1186357API
curl https://kongchang.com/api/v1/knowledge/claims/1186357MCP
get_claim(id=1186357)