Definition
The embedding dimension (or vector dimensionality) is simply the number of numeric values used to represent each item in an embedding. A 300-dimensional embedding describes every data point as a list of 300 numbers; a 1,536-dimensional one uses 1,536. It is the size of the vector space into which a model maps its inputs.
The Core Trade-Off
Dimensionality balances expressiveness against efficiency. More dimensions let an embedding encode finer distinctions, the nuanced relationships between words, or subtle visual detail, but they cost more memory, more storage, and more compute for every comparison. Too few dimensions and the model oversimplifies, blurring patterns it should keep apart. Too many and you waste resources and risk overfitting or sparse, hard-to-search vectors. The right choice depends on the data's complexity and the hardware you intend to run on.
Why It Matters for Self-Hosting
If you are building a local vector database or a retrieval-augmented system, the embedding dimension directly sets your storage footprint and query speed. Doubling the dimension roughly doubles both the disk used per vector and the work per similarity comparison. Picking a dimension that fits your machine is a practical sovereignty concern: it determines whether your AI stack runs comfortably on the hardware you control.
The embedding dimension is the size of the latent space, and comparing vectors of that dimension is usually done with cosine similarity.
In Simple Terms
The embedding dimension (or vector dimensionality) is simply the number of numeric values used to represent each item in an embedding. A 300-dimensional embedding describes…