Definition
An open-weights text embedding model optimized for long-context retrieval (8192 tokens) that utilizes Matryoshka Representation Learning to allow for flexible vector dimensions (from 64 to 768). It serves as a high-performance alternative to proprietary models in RAG pipelines, balancing high accuracy with the efficiency of variable-length vector storage.
It is an encoder-only embedding model for vectorization, not a generative LLM for chat.
"An accordion file folder that can be compressed for storage or expanded to reveal high-resolution details of a long document."
Conceptual Overview
An open-weights text embedding model optimized for long-context retrieval (8192 tokens) that utilizes Matryoshka Representation Learning to allow for flexible vector dimensions (from 64 to 768). It serves as a high-performance alternative to proprietary models in RAG pipelines, balancing high accuracy with the efficiency of variable-length vector storage.
Disambiguation
It is an encoder-only embedding model for vectorization, not a generative LLM for chat.
Visual Analog
An accordion file folder that can be compressed for storage or expanded to reveal high-resolution details of a long document.