Unverified50% confidenceFactExact time
The SE module uses global average pooling to compress the spatial information of each channel into a single scalar, then learns channel-wise recalibration weights.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSub-1-bit compression means that the average amount of information occupied per parameter is compressed to less than 1 bit.65% similarUnverifiedDisaggregated inference decouples the compute-intensive prefill stage from the memory-bandwidth-intensive decode stage, allocating them to separate hardware resource pools best suited to each for independent scaling.64% similarUnverifiedChunked KV Cache compression technology compresses Tokens into summaries within fixed-length segments using learnable gating layers, greatly reducing the VRAM pressure of long contexts, but at the same time making the position (phase) of a Token within the compression segment critically important.63% similarUnverifiedThe model compresses the cache along three dimensions: 64 query heads share a single latent, every two adjacent tokens in the encoder are compressed into one entry, and only 4 out of 40 layers retain cache while the remaining 36 layers store nothing.63% similarUnverifiedint8 quantization is a model compression technique that compresses neural network weights from 32-bit or 16-bit floating point numbers to 8-bit integers, reducing VRAM usage by approximately 50%–75%.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1028227API
curl https://kongchang.com/api/v1/knowledge/claims/1028227MCP
get_claim(id=1028227)