Reasoning on Reasoning Multi-Dataset Suite (ARB, LogiQA 2.0, ReClor, MATH)
1,580Logical Consistency Score (LCS)LoRA-SFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LoRA-SFT2026.06 | 1,580 | 10.48 | |
| ERM-SFT2026.06 | 1,420 | 9.83 | |
| CoT-Distill2026.06 | 1,390 | 10.03 | |
| IRM2026.06 | 1,040 | 8.25 | |
| V-REx2026.06 | 1,010 | 8.09 | |
| PCGrad2026.06 | 980 | 7.9 | |
| SAND-Mask2026.06 | 930 | 7.52 | |
| Pareto-GS2026.06 | 890 | 6.71 | |
| IGA (Ours)2026.06 | 310 | 4.22 | |
| GPT-4.52026.06 | 120 | 3.05 |