Back to Arena

ColPali

by illuin-tech

System Card

Organizationilluin-tech
Released2024-06
Architecturevector-rag / Multi-vector VLM page embeddings
DetailsUses PaliGemma/Qwen-VL patch outputs + linear projection + ColBERT-style late interaction to embed whole document images. Removes OCR/layout pipelines. Includes ColPali, ColQwen, ColSmol variants.
Parameters—
Domainrag-retrieval
Open SourceYes
PaperView Paper
WebsiteVisit
CodeRepository
vidoremulti-vectorcolbertvlm

Capability Profile

Benchmark Scores

5 of 14 benchmarks
Long-Context Retrieval
2/5
RULER
66.28p
NIAH
no data
LooGLE
no data
LongBench
603p
∞Bench
no data
Multi-Turn Recall
0/2
LoCoMo
no data
MemoryBank
no data
Cross-Session Memory
0/1
LongMemEval
no data
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
63.919p
HotpotQA
63.324p
Agent Task Memory
0/1
AgentBench-Mem
no data
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
1/1
RAGAS
6630p
Sources:ColPali paper (arXiv:2407.01449); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)ColPali paper (arXiv:2407.01449); evaluated on LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding (Tsinghua KEG, 2308)ColPali paper (arXiv:2407.01449); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)ColPali paper (arXiv:2407.01449); evaluated on RAGAS: Automated Evaluation of Retrieval-Augmented Generation (Exploding Gradients, 2309)ColPali paper (arXiv:2407.01449); evaluated on RULER: What's the Real Context Size of Your Long-Context Language Models (NVIDIA, 2404)