Document Question Answering on DocVQA
97.87ANLSBaseline
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BaselineBackbone=LLaVA-OneVision-1.5-8B-Instruct2026.02 | 97.87 | — | |
| VisionSelectorRetain Tokens=25%2026.02 | 94.08 | — | |
| IDPrunerRetain Tokens=25%2026.02 | 91.82 | — | |
| VanillaBackbone=Qwen2.5-VL-7B, Token Retention Rate=100%2025.12 | 91 | — | |
| ERNIE-LayoutlargeFine-tuning set=train+dev, Context=leaderboard2022.10 | 88.41 | — | |
| TILTlargeFine-tuning set=-2022.10 | 87.05 | — | |
| LayoutLMv2largeFine-tuning set=train + dev2022.10 | 85.29 | 0.082 | |
| ERNIE-LayoutlargeFine-tuning set=train+dev2022.10 | 84.86 | 0.1534 | |
| DARTRetain Tokens=25%2026.02 | 84.77 | — | |
| StructuralLMlargeFine-tuning set=-2022.10 | 83.49 | — | |
| LayoutLMv2largeFine-tuning set=train2022.10 | 83.48 | 0.0639 | |
| ERNIE-LayoutlargeFine-tuning set=train2022.10 | 83.21 | 0.1369 | |
| DivPruneRetain Tokens=25%2026.02 | 81.53 | — | |
| D²PrunerBackbone=Qwen2.5-VL-7B, Token Retention Rate=25%2025.12 | 81.14 | — | |
| UniLMv2largeFine-tuning set=train2022.10 | 77.09 | — | |
| SCOPERetain Tokens=25%2026.02 | 75.68 | — | |
| DivPruneBackbone=Qwen2.5-VL-7B, Token Retention Rate=25%2025.12 | 73.88 | — | |
| XDocPre-training Steps=1M, Pre-train (P)=true, Pre-train (D)=true, Pre-train (W)=true2022.10 | 72.7 | — | |
| XDocPre-training Steps=500K, Pre-train (P)=true, Pre-train (D)=true, Pre-train (W)=true2022.10 | 72.6 | — | |
| LayoutLMlargeFine-tuning set=train2022.10 | 72.59 | — | |
| VisionSelectorRetain Tokens=10%2026.02 | 71.31 | — | |
| FastVBackbone=Qwen2.5-VL-7B, Token Retention Rate=25%2025.12 | 71.26 | — | |
| IDPrunerRetain Tokens=10%2026.02 | 70.43 | — | |
| ROBERTalargeFine-tuning set=train2022.10 | 69.52 | — | |
| XDocPre-training Steps=100K, Pre-train (D)=true, Pre-train (W)=true2022.10 | 69.5 | — | |
| XDocPre-training Steps=100K, Pre-train (D)=true2022.10 | 69.4 | — | |
| LayoutLMPre-train (D)=true2022.10 | 69.2 | — | |
| XDocPre-training Steps=100K, Pre-train (P)=true, Pre-train (D)=true2022.10 | 69.1 | — | |
| XDocPre-training Steps=100K, Pre-train (P)=true, Pre-train (D)=true, Pre-train (W)=true2022.10 | 68.3 | — | |
| BERTlargeFine-tuning set=train2022.10 | 67.68 | — | |
| UNITBase LVLM=Llava-Next, Resolution Strategy=high-resolution grid slicing, Resolution=4482024.09 | 65.5 | — | |
| XDocPre-training Steps=100K, Pre-train (P)=true, Pre-train (W)=true2022.10 | 64.7 | — | |
| XDocPre-training Steps=100K, Pre-train (P)=true2022.10 | 64.5 | — | |
| XDocPre-training Steps=100K, Pre-train (W)=true2022.10 | 63.6 | — | |
| SigLIPBase LVLM=Llava-Next, Resolution Strategy=high-resolution grid slicing, Resolution=3842024.09 | 62 | — | |
| VisPrunerRetain Tokens=25%2026.02 | 61 | — | |
| D²PrunerBackbone=Qwen2.5-VL-7B, Token Retention Rate=10%2025.12 | 57.54 | — | |
| CLIP-LBase LVLM=Llava-Next, Resolution Strategy=high-resolution grid slicing, Resolution=3362024.09 | 57.2 | — | |
| DivPruneRetain Tokens=10%2026.02 | 56.57 | — | |
| FastVRetain Tokens=25%2026.02 | 56.53 | — | |
| DARTBackbone=Qwen2.5-VL-7B, Token Retention Rate=25%2025.12 | 52.99 | — | |
| DARTRetain Tokens=10%2026.02 | 52.77 | — | |
| DivPruneBackbone=Qwen2.5-VL-7B, Token Retention Rate=10%2025.12 | 49.34 | — | |
| SCOPERetain Tokens=10%2026.02 | 47.65 | — | |
| FastVBackbone=Qwen2.5-VL-7B, Token Retention Rate=10%2025.12 | 36.85 | — | |
| VisPrunerRetain Tokens=10%2026.02 | 35.18 | — | |
| DARTBackbone=Qwen2.5-VL-7B, Token Retention Rate=10%2025.12 | 30.1 | — | |
| FastVRetain Tokens=10%2026.02 | 29.77 | — | |
| VisionZipRetain Tokens=25%2026.02 | 23.69 | — | |
| HiPruneRetain Tokens=25%2026.02 | 21.01 | — | |
| VisionZipRetain Tokens=10%2026.02 | 13.2 | — | |
| HiPruneRetain Tokens=10%2026.02 | 12.86 | — | |
| MMaDA+VRCDL=384, FR=0.52026.05 | 12.38 | — | |
| MMaDA+VRCDL=192, FR=0.52026.05 | 11.88 | — | |
| MMaDA+VRCDL=384, FR=0.252026.05 | 11.74 | — | |
| MMaDAL=384, FR=0.52026.05 | 11.64 | — | |
| MMaDA+VRCDL=192, FR=0.252026.05 | 11.38 | — | |
| MMaDAL=384, FR=0.252026.05 | 11.23 | — | |
| MMaDAL=192, FR=0.52026.05 | 11.19 | — | |
| MMaDAL=192, FR=0.252026.05 | 10.18 | — | |
| MMaDA+VRCDL=384, FR=0.1252026.05 | 9.58 | — | |
| MMaDAL=384, FR=0.1252026.05 | 9.27 | — | |
| MMaDA+VRCDL=192, FR=0.1252026.05 | 9.09 | — | |
| MMaDAL=192, FR=0.1252026.05 | 8.49 | — |