stanford-futuredata/ColBERT
A neural retrieval model that encodes passages and queries into token-level embedding matrices and uses late interaction to efficiently compute fine-grained similarity at scale.
Not currently ranked — collecting fresh signals.
star history
ColBERT is a fast and accurate retrieval system built on BERT that encodes each passage into a matrix of token-level embeddings. At search time, it embeds queries into similar matrices and employs scalable vector-similarity MaxSim operators to efficiently find contextually matching passages. This late-interaction approach enables it to surpass single-vector representation models in quality while maintaining the ability to scale to large document collections.
Frequently asked
- What is stanford-futuredata/ColBERT?
- A neural retrieval model that encodes passages and queries into token-level embedding matrices and uses late interaction to efficiently compute fine-grained similarity at scale.
- Is ColBERT open source?
- Yes — stanford-futuredata/ColBERT is open source, released under the MIT license.
- What language is ColBERT written in?
- stanford-futuredata/ColBERT is primarily written in Python.
- How popular is ColBERT?
- stanford-futuredata/ColBERT has 3.9k stars on GitHub.
- Where can I find ColBERT?
- stanford-futuredata/ColBERT is on GitHub at https://github.com/stanford-futuredata/ColBERT.