Premise Selection on Premise Selection (val)
26.4RecallGPT-5.5
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-5.5k=10, H=62026.06 | 26.4 | 60 | |
| Qwen3-8B SDPO + same-paper graph augmentationk=10, H=6, Training Strategy=SDPO, Graph Augmentation Strategy=same-paper2026.06 | 22.7 | 140 | |
| Gemini 3.1 Prok=10, H=62026.06 | 17 | 60 | |
| Qwen3-8B SDPO + cross-paper graph augmentationk=10, H=6, Training Strategy=SDPO, Graph Augmentation Strategy=cross-paper2026.06 | 13.4 | 115 | |
| Qwen3-8B SDPOk=10, H=6, Training Strategy=SDPO, Graph Augmentation Strategy=None2026.06 | 7.9 | 60 | |
| Qwen3-8B basek=10, H=6, Training Strategy=None2026.06 | 4 | 60 |