Unverified50% confidenceFactExact time
IVFFlat clusters the vector space into multiple Voronoi cells; on insertion it only computes the owning cluster and appends, with no global graph restructuring, so write throughput is higher, but recall is usually lower than HNSW.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedHNSW构建多层图结构，高层稀疏覆盖大范围，低层密集精确定位，在亿级向量规模下仍能实现毫秒级响应59% similarVerifiedLoRA decomposes the weight update matrix ΔW into the product of two small matrices A and B whose ranks are much lower than the original matrix, reducing parameter count to as little as one ten-thousandth of the original or even less.59% similarUnverifiedIn large-scale GPU matrix operations, the associative law of floating-point addition does not strictly hold, and different parallel reduction orders, batch sizes, and memory scheduling can cause minor discrepancies in logits.59% similarVerifiedHNSW reduces search complexity to approximately O(log n) by constructing a multi-layer graph index, while IVF partitions the vector space via K-means clustering to achieve orders-of-magnitude speed improvements at the cost of approximately 5–10% recall.59% similarUnverifiedLow-rank decomposition and quantization compression are orthogonally complementary dimensions; quantizing the smaller matrices A and B separately results in smaller errors, and the combined compression ratio can reach several times that of either method alone58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1057163API
curl https://kongchang.com/api/v1/knowledge/claims/1057163MCP
get_claim(id=1057163)