[KongchangAI]
Concept嵌入表 / embedding 表

embedding table

A data structure in recommendation systems that stores dense vectors corresponding to discrete features. Sparse integer indices are mapped to trainable vectors, and the table can reach billions of rows, typically sharded by row across multiple GPUs.

Timeline (last 90 days)

Oct 12

The embedding table size of recommendation systems is often on the same order as the feature cardinality, reaching billions of rows, which cannot fit in a single GPU's memory, so it must be split by row across multiple GPUs.

Unverified50%

All Facts (1)

Source Articles