Unverified50% confidenceSolutionExact time
The project feeds a deterministic, normalized inverse-lag history vector P_t into the recurrent unit, weighted by harmonic numbers, causing historical information to decay at a rate of 1/k.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe project efficiently computes the history vector P_t via zero-padded FFT linear convolution, keeping overall complexity at O(T log T).70% similarUnverifiedAutoregressive models use KV Cache to cache the Key and Value matrices of historical tokens, with memory usage growing linearly with sequence length.65% similarUnverifiedIn standard Transformer inference, generating each new token requires accessing the key-value vectors of all historical tokens, and the memory footprint of the KV cache grows linearly with sequence length.64% similarUnverifiedAutoregressive generative models sample the next token at each step based on the probability distribution of the preceding context; when a repetitive pattern gains a slightly higher probability, it forms a positive feedback loop that leads to repetition degeneration.64% similarVerifiedThe autoregressive generation process of reasoning models is strictly sequential; KV Cache VRAM usage grows linearly with the length of the thinking token sequence, making memory bandwidth the performance bottleneck.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/902244API
curl https://kongchang.com/api/v1/knowledge/claims/902244MCP
get_claim(id=902244)