Visual Document Retrieval on ViDoRe 10 datasets v1
90.6NDCG@5Tomoro-8B
Evaluation results
| Method | Links | |
|---|---|---|
| Tomoro-8BBackbone=ColQwen3-8B, Params=8.0B, Scoring=MaxSim2026.03 | 90.6 | |
| Tomoro-4BBackbone=ColQwen3-4B, Params=4.0B, Scoring=MaxSim2026.03 | 90.2 | |
| ColNomic-7BBackbone=ColQwen2.5-7B, Params=7.0B, Scoring=MaxSim2026.03 | 89.8 | |
| DSE-Qwen2Backbone=Qwen2-VL-2B, Params=2.0B, Scoring=Cosine2026.03 | 85.1 | |
| Qwen3-VL-EmbBackbone=Qwen3-VL-2B, Params=2.0B, Scoring=Cosine2026.03 | 84.3 | |
| ColPaliBackbone=PaliGemma-3B, Params=3.0B, Scoring=MaxSim2026.03 | 84.2 | |
| NanoVDR-LBackbone=ModernBERT-base, Params=151M, Scoring=Cosine2026.03 | 82.4 | |
| NanoVDR-SBackbone=DistilBERT, Params=69M, Scoring=Cosine2026.03 | 82.2 | |
| NanoVDR-S-MultiBackbone=DistilBERT, Params=69M, Scoring=Cosine2026.03 | 82.2 | |
| NanoVDR-MBackbone=BERT-base, Params=112M, Scoring=Cosine2026.03 | 82.1 | |
| ColModernVBertBackbone=ModernBERT-base, Params=250M, Scoring=MaxSim2026.03 | 76.7 | |
| JinaCLIPBackbone=CLIP-ViT-L, Params=400M, Scoring=Cosine2026.03 | 53.7 | |
| SigLIP2Backbone=So400m, Params=400M, Scoring=Cosine2026.03 | 44.6 | |
| BiModernVBertBackbone=ModernBERT-base, Params=250M, Scoring=Cosine2026.03 | 37.4 |