Visual Question Answering on SeedBench Avg
77.96AccuracyE2E
Evaluation Results
| Method | Links | |
|---|---|---|
| E2EVLM=Qwen2.5-VL, Setting=RL2025.12 | 77.96 | |
| E2EVLM=Qwen2.5-VL, Setting=base + CoT2025.12 | 77.6 | |
| VISTAVLM=Qwen2.5-VL, Setting=RL2025.12 | 76.4 | |
| E2EVLM=Qwen2.5-VL, Setting=SFT2025.12 | 75.67 | |
| E2EVLM=Qwen2.5-VL, Setting=base2025.12 | 75.31 | |
| VISTAVLM=Llama3.2-Vision, Setting=RL2025.12 | 75.21 | |
| E2EVLM=Llama3.2-Vision, Setting=base + CoT2025.12 | 74.77 | |
| E2EVLM=Llama3.2-Vision, Setting=base2025.12 | 73.34 | |
| VISTAVLM=Qwen2.5-VL, Setting=base2025.12 | 68.86 | |
| E2EVLM=Llama3.2-Vision, Setting=SFT2025.12 | 68.18 | |
| VISTAVLM=Llama3.2-Vision, Setting=base2025.12 | 66.19 |