Unverified50% confidenceSolutionExact time
Focus67Dumby's perceptual-diff caching mechanism scales the in-memory screenshot to 512x288 (JPEG quality 65) and computes a 16x9 grayscale perceptual signature; if the pixel change is below 3% and the window title is the same, it hits the local session cache, achieving 0 ms latency and 0 GPU calls.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSHADOW 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.73% similarUnverifiedChecking at low resolution (such as a 600px thumbnail) is more effective than at high resolution (a 4000px preview), as it can reveal residual gray bands and saves memory. On a 1 vCPU, 2GB memory machine, the author reduced check rendering from 300 dpi to 60 dpi, and peak memory dropped substantially, no longer triggering swap.73% similarUnverifiedIn 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.71% similarUnverifiedAdopting sparse attention can compress the KV cache to approximately 1/4, supporting approximately 4 times the context length of full attention under the same memory budget.69% similarUnverifiedWhen the 120B coordinator is resident (about 77GB), only about 3GB of memory is free in the system; in a multi-image run the first image completes and the remaining images time out at the 300-second limit and fall back to placeholders.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1055099API
curl https://kongchang.com/api/v1/knowledge/claims/1055099MCP
get_claim(id=1055099)