Visual Question Answering on AI2D (Exact Match)
82.48EMBaseline
Evaluation Results
| Method | Links | |
|---|---|---|
| BaselineCompression Ratio=0%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 82.48 | |
| BaselineBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=100%, Compression Ratio=0%2026.02 | 82.48 | |
| PaLI-XOCR pipeline input=true, Resolution=812x8122023.10 | 81.4 | |
| PaLI-XOCR pipeline input=false, Resolution=812x8122023.10 | 81.2 | |
| IDPrunerCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 80.51 | |
| IDPrunerBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=25%, Compression Ratio=75%2026.02 | 80.51 | |
| VisionSelectorCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 79.6 | |
| VisionSelectorBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=25%, Compression Ratio=75%2026.02 | 79.6 | |
| SCOPECompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 78.92 | |
| SCOPEBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=25%, Compression Ratio=75%2026.02 | 78.92 | |
| DivPruneCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 77.98 | |
| DivPruneBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=25%, Compression Ratio=75%2026.02 | 77.98 | |
| VisPrunerCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 77.62 | |
| VisPrunerBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=25%, Compression Ratio=75%2026.02 | 77.62 | |
| HiPruneCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 77.49 | |
| HiPruneBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=25%, Compression Ratio=75%2026.02 | 77.49 | |
| VisionZipCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 77.4 | |
| VisionZipBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=25%, Compression Ratio=75%2026.02 | 77.4 | |
| FastVCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 75.68 | |
| FastVBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=25%, Compression Ratio=75%2026.02 | 75.68 | |
| PaLI-3OCR pipeline input=true, Resolution=812x8122023.10 | 75.2 | |
| PaLI-3OCR pipeline input=false, Resolution=812x8122023.10 | 75.2 | |
| IDPrunerCompression Ratio=90%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 75.16 | |
| IDPrunerBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=10%, Compression Ratio=90%2026.02 | 75.16 | |
| VisionSelectorCompression Ratio=90%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 74.81 | |
| VisionSelectorBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=10%, Compression Ratio=90%2026.02 | 74.81 | |
| DARTCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 74.35 | |
| DARTBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=25%, Compression Ratio=75%2026.02 | 74.35 | |
| SCOPECompression Ratio=90%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 71.63 | |
| SCOPEBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=10%, Compression Ratio=90%2026.02 | 71.63 | |
| VisionZipCompression Ratio=90%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 70.6 | |
| VisionZipBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=10%, Compression Ratio=90%2026.02 | 70.6 | |
| DivPruneCompression Ratio=90%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 70.11 | |
| DivPruneBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=10%, Compression Ratio=90%2026.02 | 70.11 | |
| VisPrunerCompression Ratio=90%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 69.88 | |
| VisPrunerBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=10%, Compression Ratio=90%2026.02 | 69.88 | |
| HiPruneCompression Ratio=90%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 69.82 | |
| HiPruneBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=10%, Compression Ratio=90%2026.02 | 69.82 | |
| DARTCompression Ratio=90%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 67.88 | |
| DARTBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=10%, Compression Ratio=90%2026.02 | 67.88 | |
| FastVCompression Ratio=90%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 67.23 | |
| FastVBase Model=Qwen-2.5-VL-7B-Instruct, Retained Tokens=10%, Compression Ratio=90%2026.02 | 67.23 |