Tiny vision-language models for visual document retrieval, built on UltraBERT.
AI & ML interests
Late interaction and multi-vector retrieval: ColBERT-style text retrievers, ColPali-style visual document retrieval, MaxSim scoring, token pruning and index compression, efficient training recipes, and open benchmarks.
Recent Activity
Organization Card
Late Interaction
An open community for training, evaluating, and sharing late interaction retrieval models: ColBERT-style text retrievers, ColPali-style visual document retrievers, and other multi-vector approaches.
What we work on
- Training recipes: reproducible configs for pretraining, distillation, and fine-tuning multi-vector models
- Efficiency: token pruning, quantization, and index compression
- Multimodal retrieval: documents, images, and long-context inputs
- Evaluation: honest benchmarks with full eval code released alongside results
Naming conventions
To keep the org browsable, please follow these patterns:
| Type | Pattern | Example |
|---|---|---|
| Models | {backbone}-{method}-{variant} |
modernbert-base-colbert-msmarco |
| Datasets | {source}-{task}-{format} |
msmarco-triplets-hardnegs |
| Experiments | prefix with exp- |
exp-pruning-ablation-v2 |
Every model should include a card with the training data, hyperparameters, eval results, and the code or config used to train it.
Contributing
- Ask an admin to add you to the org
- Discuss larger efforts in the Community tab before training
- Push checkpoints to your personal namespace first, then move them here once evaluated
Related work
We build on and are grateful for ColBERT, PLAID, ColPali, and PyLate.
models 0
None public yet
datasets 0
None public yet