Abstention on ABSTAINEQA HM3D-380
71.1F1 ScoreSEMANTIC FLIP
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SEMANTIC FLIPParams=7B + small head2026.06 | 71.1 | 66.84 | 81.58 | 52.11 | |
| Qwen-32B-CoarseParams=32B2026.06 | 67.46 | 56.84 | 89.47 | 24.21 | |
| Qwen-32B-FineParams=32B2026.06 | 67.41 | 53.95 | 95.26 | 12.63 | |
| Qwen-32B-Fine + Naive CoTParams=32B2026.06 | 30.99 | 56.54 | 19.6 | 93.44 |