# illuin-tech/colpali

The code used to train and run inference with the ColVision models, e.g. ColPali, ColQwen2, and ColSmol.

Repository: https://github.com/illuin-tech/colpali
Canonical: https://ross.abutalabs.com/products/colpali
Homepage: https://huggingface.co/vidore
Language: Python
License: MIT
License Family: permissive
Topics: colpali, information-retrieval, retrieval-augmented-generation, vision-language-model, colqwen2, colsmol
Last push: 2026-08-24T14:35:20+00:00

## Health v2 (maintenance only)
Score: 91/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 99, release rhythm 99, longevity 57
- inputs: {"age_days": 804, "days_push": 9, "days_rel": 11, "gap_med": 26, "n_releases_24m": 20}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2798, forks 261 (observed 2026-08-28T04:07:22.497017+00:00)

## What it is
ColPali Engine is the Python library for training and running inference with ColVision visual document retrieval models such as ColPali, ColQwen2, and ColSmol. It uses vision-language models to build multi-vector embeddings of document pages, replacing brittle OCR and layout-parsing pipelines. The package is now deprecated in favor of Sentence Transformers' MultiVectorEncoder.

## Use cases
- retrieve documents from pdfs using vision language models
- train a ColPali-style visual document retriever
- index scanned document pages without OCR pipelines
- run inference with ColQwen2 or ColSmol models
- build a RAG pipeline over visually complex documents
- reproduce results from the ColPali paper and ViDoRe benchmark

## When to choose
- you need to reproduce ColPali/ColVision research results or use existing trained checkpoints
- your documents are visually complex (tables, charts, layouts) and OCR pipelines fail
- you want multi-vector late-interaction retrieval over document page images

## When to avoid
- you are starting a new production project - use Sentence Transformers v6 MultiVectorEncoder instead
- you only need plain text embedding retrieval without visual content
- you cannot run GPU inference for vision-language models

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, rag, search-engine, llm-training, nlp
- domain: machine-learning, computer-vision, pdf
- platform: python
- tags: vision-language-models, document-retrieval, colpali, colqwen2, multi-vector-embeddings, colbert, vidore-benchmark, deprecated, retrieval-augmented-generation, search, gpu

## Member repositories
- illuin-tech/colpali (main) score 91

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:22.497017+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T08:14:38.105534+00:00, confidence not recorded.
  - readme: https://github.com/illuin-tech/colpali (fetched 2026-08-28T04:07:22.497017+00:00, sha 205af39b27a4)
  - homepage: https://huggingface.co/vidore (fetched 2026-08-29T09:54:49.884280+00:00, sha 29400aa6457b)
  - site_page: https://huggingface.co/docs (fetched 2026-08-29T09:54:49.894258+00:00, sha bdec26667b98)
  - site_page: https://huggingface.co/docs/hub/organizations-cards (fetched 2026-08-29T09:54:49.898119+00:00, sha 635040b5f704)
  - site_page: https://huggingface.co/pricing (fetched 2026-08-29T09:54:49.896106+00:00, sha de6b7a178be5)
  - site_page: https://huggingface.co/huggingface (fetched 2026-08-29T09:54:49.899643+00:00, sha 740804c66413)
- Data as of 2026-08-30T08:39:29.467469+00:00.