ML//RAG//embedding model

An embedding model is a dedicated model trained to produce dense vector representations optimized for retrieval. SentenceTransformers, OpenAI embeddings, Nomic Embed and Jina v3 are each tuned for semantic similarity, and they are evaluated on the MTEB leaderboard, the de facto benchmark despite its known overfitting.

An embedding model is a dedicated model trained to produce dense vector representations optimized for retrieval. SentenceTransformers, OpenAI embeddings, Nomic Embed and Jina v3 are each tuned for semantic similarity, and they are evaluated on the MTEB leaderboard, the de facto benchmark despite its known overfitting.

Matryoshka embeddings are trained once and then truncated to any dimension, which gives a flexible tradeoff between quality and cost.