Vision-Language Question Answering on ScienceQA
92.4AccuracyLoRA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LoRAModel=Qwen2.5-VL-3B, Optimization=FlatPO2026.06 | 92.4 | — | |
| LoRAModel=Qwen2.5-VL-3B, Optimization=GAM2026.06 | 91.8 | — | |
| LoRAModel=Qwen2.5-VL-3B, Optimization=SAM2026.06 | 91.7 | — | |
| Prefix TuningModel=Qwen2.5-VL-3B, Optimization=FlatPO2026.06 | 90.9 | — | |
| LoRAModel=Qwen2.5-VL-3B, Optimization=Base2026.06 | 90.6 | — | |
| Prefix TuningModel=Qwen2.5-VL-3B, Optimization=GAM2026.06 | 89.8 | — | |
| Prefix TuningModel=Qwen2.5-VL-3B, Optimization=SAM2026.06 | 87.8 | — | |
| Prefix TuningModel=Qwen2.5-VL-3B, Optimization=Base2026.06 | 86.2 | — | |
| FP16 (Full Precision)W/A Bits=16/162026.06 | 79.39 | 11.79 | |
| Qwen2.5-Omni (Ours)W/A Bits=4/42026.06 | 76.63 | 6.07 | |
| QLoRAW/A Bits=4/162026.06 | 75.88 | 5.54 | |
| PoMQ-ViTW/A Bits=4/42026.06 | 75.36 | 6.37 | |
| QVLMW/A Bits=4/42026.06 | 73.85 | 5.03 | |
| Naive W4A4W/A Bits=4/42026.06 | 73.17 | 6.02 |