Short-form single-page QA on HotpotQA (official)
34.3Exact Match (EM)AGAR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AGARBackbone=Qwen3-VL-8B-Instruct2026.06 | 34.3 | 46.9 | 291 | |
| VEA-ConBackbone=Qwen3-VL-8B-Instruct2026.06 | 30.9 | 41.6 | 268 | |
| RandomBackbone=Qwen3-VL-8B-Instruct2026.06 | 30.4 | 41.8 | 301 | |
| UniformBackbone=Qwen3-VL-8B-Instruct2026.06 | 30.3 | 41.4 | 292 | |
| AGLABackbone=Qwen3-VL-8B-Instruct2026.06 | 30.2 | 41.2 | 268 | |
| VEA-BrBackbone=Qwen3-VL-8B-Instruct2026.06 | 30 | 40.5 | 268 | |
| Plain VQABackbone=Qwen3-VL-8B-Instruct2026.06 | 29.2 | 39.7 | 268 | |
| VARBackbone=Qwen3-VL-8B-Instruct2026.06 | 26.8 | 38.3 | 268 | |
| CGRBackbone=Qwen3-VL-8B-Instruct2026.06 | 22.6 | 29.5 | 268 |