Embedding Models

Embedding models transform text into high-dimensional vectors that represent semantic meaning. The quality of these embeddings is crucial for effective retrieval.


Key Considerations:

  • Must support both the query and the document chunks.

  • Should be trained or fine-tuned on similar types of text.

  • Trade-off: Smaller models are faster; larger models are more accurate.

Common Embedding Models:

Model NameTypeNotes
intfloat/e5-small-v2DenseLightweight and fast; good baseline
intfloat/e5-large-v2DenseMore accurate; used for robust retrieval
BAAI/bge-small-enDenseHigh-quality general-purpose model
BAAI/bge-large-en-v1.5DenseOne of the most accurate public models
GTE-base / GTE-smallDenseLightweight alternatives by Alibaba
all-mpnet-base-v2DenseFrom SentenceTransformers; widely used baseline
Jina Embeddings v2DenseOptimized for semantic search

Choosing a Model:

  • for production use we went with nomic-embed-text .