Glossary

Vector

An ordered list of numbers that represents a point or direction in continuous space—embeddings and activations are vectors at different stages of the model.

What It Is

A vector is a fixed-length sequence of real numbers. In language models, token embeddings are vectors of model width, and attention layers read and write vectors at every position along the sequence.

Why It Matters

Papers and code often say "vector" when they mean a 1-D tensor: one row from an embedding table, a query vector, or a hidden state. Knowing that vectors are the building blocks inside tensors helps you follow shape notation from embeddings through attention.

Simple Example

A 768-dimensional embedding for one token is a vector of length 768. Stacking many positions produces a matrix or higher-rank tensor, but each row is still a vector in hidden space.

Common Confusions

A vector is not the same as a sparse one-hot encoding: learned embeddings are dense vectors with values in every dimension. "Dense vector" usually contrasts with sparse bag-of-words features, not with tensors in general.

Tags

References