# colbert

> Browse 9 items tagged with colbert.

## Items

- [ConstBERT](https://vectordb.ever.works/items/constbert) — Novel approach to reduce storage footprint of multi-vector retrieval by encoding each document with a fixed, smaller set of learned embeddings. Reduces index sizes by over 50% compared to ColBERT while retaining most effectiveness.
- [FastPLAID](https://vectordb.ever.works/items/fastplaid) — Optimized implementation of PLAID index for fast ColBERT retrieval, providing 10x storage compression and sub-200ms latency. Default index backend for PyLate library, enabling efficient multi-vector late interaction retrieval.
- [PyLate](https://vectordb.ever.works/items/pylate) — Library built on Sentence Transformers for flexible training, inference, and retrieval with state-of-the-art ColBERT models. Features FastPLAID index for efficient multi-vector late interaction retrieval with 10x storage compression and sub-200ms latency.
- [Multi-Vector Embeddings](https://vectordb.ever.works/items/multi-vector-embeddings) — Embedding approach where documents/images are represented by multiple vectors (one per token/patch) rather than a single vector, enabling fine-grained semantic matching.
- [Late Interaction Retrieval](https://vectordb.ever.works/items/late-interaction-retrieval) — A retrieval paradigm where query and document encodings are kept separate until a late interaction stage, enabling more expressive and efficient similarity computations. Pioneered by ColBERT and extended by ColPali and ColQwen, this approach maintains fine-grained representations while enabling fast retrieval.
- [Late Interaction](https://vectordb.ever.works/items/late-interaction) — Retrieval paradigm where query and document tokens are encoded separately and interactions computed at search time, combining efficiency of bi-encoders with expressiveness of cross-encoders.
- [MaxSim Operator](https://vectordb.ever.works/items/maxsim-operator) — Scoring function used in late interaction models like ColBERT that computes query-document relevance by finding maximum similarity between each query token and document tokens, then summing.
- [Jina ColBERT v2](https://vectordb.ever.works/items/jina-colbert-v2) — Groundbreaking multilingual information retrieval model supporting 89 languages with token-level embeddings and late interaction. Features Matryoshka embeddings for flexible efficiency-precision tradeoffs and 8192 token input context.
- [MaxSim](https://vectordb.ever.works/items/maxsim) — Maximum Similarity late interaction function introduced by ColBERT for ranking. Calculates cosine similarity between query and document token embeddings, keeping maximum score per query token for highly effective long-document retrieval.

---

_Canonical page: https://vectordb.ever.works/tags/colbert_