Embedding

Appears in 3 papers · 5 tutorials

A dense, low-dimensional vector representation of something — a word,

As used in Paper 05 — Efficient Estimation of Word Representations in Vector Space (Word2Vec) →

A dense, low-dimensional vector representation of something — a word,

As used in Paper 06 — Sequence to Sequence Learning with Neural Networks →

A dense, low-dimensional vector representation of a word. Introduced

As used in Paper 20 — Gemini: A Family of Highly Capable Multimodal Models →

A learned vector representation of a token (text word, image patch, audio feature, etc.). For Gemini, all embeddings are d_model = 2048 dimensions, ensuring uniform treatment.

As used in AI Agents →

A way to turn text into numbers (a vector) that capture meaning, so similar ideas have similar numbers.

As used in Retrieval-Augmented Generation (RAG) →

A list of numbers representing the meaning of a piece of text; similar meanings → similar vectors.

As used in Vector Databases →

A list of numbers representing the meaning of something (text, image, audio). Similar meanings → similar vectors. The atom of everything.

As used in Fine-Tuning & Model Customization →

A list of numbers (a vector) representing a token's "meaning" as coordinates in a high-dimensional space; similar meanings sit close together. (M01)

As used in AI Production Engineering →

A list of numbers representing the meaning of a piece of text, so similar meanings land near each other. (Mod 1)