Unverified50% confidenceSolutionExact time
SHADOW writes attention states to disk at 1-bit precision, consuming 288 bytes per token, with the index occupying only 22 bytes per token, record lookup taking approximately 1 microsecond, and attention states being reinjected into the model in approximately 0.03 ms, without using a vector database or embedding model.
1
Sources
50%
Confidence
Long-term
Relevance
9/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn a video memory experiment, Gemma 3 4B watched a ten-minute short film, captured a frame every two seconds, and wrote 298 descriptions; SHADOW retained these moments as disk memory and scored 55–56/60 on a laptop with approximately 42 MB of memory.73% similarVerifiedFor a typical 7B model with 32 layers and 32 attention heads, storing the KV Cache for each token requires approximately 0.5 MB of VRAM, so generating 1,000 tokens requires an additional ~500 MB of VRAM71% similarUnverifiedThe attention matrix size required to process a 128K token sequence is 256 times that of an 8K token sequence.68% similarUnverifiedSHADOW does not use a trained embedding layer; instead, it represents a vocabulary of 73,880 tokens using fixed 512-bit fingerprints, the frozen table is only 4.7 MB, and new tokens can be added at any time without retraining.66% similarUnverifiedIn large model API call scenarios, cache hit rate refers to the reuse ratio of the KV Cache (key-value cache), where the model can directly read the attention key-value pairs of already-processed tokens from the cache.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/922676API
curl https://kongchang.com/api/v1/knowledge/claims/922676MCP
get_claim(id=922676)