Multimodal Understanding on lmms-eval Zero-Shot Suite (MME, VQA^T, POPE, RWQA, MMB, MMS)
72.6Average ScoreFourier-Qwen2.5-VL-3B
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Fourier-Qwen2.5-VL-3BModel Series=Qwen2.5-VL-3B, Method=Fourier-, Average vision tokens (#I)=2362025.08 | 72.6 | 2,110.8 | 76.1 | 87.8 | 64.3 | 76.8 | 55.3 | |
| Qwen2.5-VL-3BModel Series=Qwen2.5-VL-3B, Method=Vanilla, Average vision tokens (#I)=5532025.08 | 71.1 | 2,138.3 | 77.6 | 87.2 | 59.6 | 77.8 | 56.2 | |
| Qwen2-VL-2BModel Series=Qwen2-VL-2B, Method=Vanilla, Average vision tokens (#I)=5532025.08 | 66.8 | 1,894.4 | 78.2 | 86.1 | 61.4 | 72.3 | 44.2 | |
| Fourier-Qwen2-VL-2BModel Series=Qwen2-VL-2B, Method=Fourier-, Average vision tokens (#I)=2362025.08 | 65.7 | 1,917.1 | 75.4 | 86.5 | 61.6 | 72.6 | 43.3 |