Unverified50% confidenceFactExact time
Sequence parallelism splits the inputs of operators such as LayerNorm and Dropout along the sequence dimension, causing activation memory to decrease linearly with the degree of parallelism
1
Sources
50%
Confidence
Long-term
Relevance
10/6/2026
First Seen
Sources
Related Entities
Related Claims
Unverified基于序列号的槽位机制中，环形缓冲区每个槽位携带独立序列号，将全局头尾指针写争用分散至各槽位，减少缓存行冲突69% similarUnverifiedFor graphs with small integer edge weights, Dial's Algorithm (a bucket-sort approach) can reduce the complexity to O(m + C), where C is the maximum edge weight.66% similarUnverifiedWhen there is no ordering relationship between the rows of matrix data (swapping rows does not affect the classification result), the data is essentially a Set rather than a Sequence.65% similarUnverified倒排乘积量化（IVFPQ）在PQ基础上引入倒排索引结构，通过聚类将向量库分区，将检索复杂度从线性降低至亚线性64% similarUnverified格是由一组线性无关的基向量通过整数线性组合所张成的离散点集合63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/982626API
curl https://kongchang.com/api/v1/knowledge/claims/982626MCP
get_claim(id=982626)