Multimodal Question Answering on MMBench CN (test)
88.9AccuracyVanilla
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VanillaBackbone=InternVL3-38B, ViT token retention rate=100%, LLM token retention rate=100%2025.12 | 88.9 | 100 | 1 | |
| DART (ViT)Backbone=InternVL3-38B, ViT token retention rate=40%, LLM token retention rate=40%2025.12 | 85.4 | 88.1 | 0.639 | |
| ToFuBackbone=InternVL3-38B, ViT token retention rate=40%, LLM token retention rate=40%2025.12 | 83.5 | 83.4 | 0.619 | |
| IPCV+FastVBackbone=InternVL3-38B, ViT token retention rate=40%, LLM token retention rate=40%2025.12 | 83.2 | 89.9 | 0.644 | |
| ToMeBackbone=InternVL3-38B, ViT token retention rate=40%, LLM token retention rate=40%2025.12 | 80.9 | 82.7 | 0.617 | |
| FastVBackbone=InternVL3-38B, ViT token retention rate=100%, LLM token retention rate=5%2025.12 | 77 | 78.6 | 0.707 | |
| SparseVLMBackbone=InternVL3-38B, ViT token retention rate=100%, LLM token retention rate=5%2025.12 | 73.5 | 75.1 | 0.734 | |
| DARTBackbone=InternVL3-38B, ViT token retention rate=100%, LLM token retention rate=5%2025.12 | 70.5 | 73.6 | 0.715 | |
| V^2DropBackbone=InternVL3-38B, ViT token retention rate=100%, LLM token retention rate=5%2025.12 | 59.1 | 57.1 | 0.705 |