Natural Language Inference on MultiNLI reconstructed with controlled shortcut injection (test)
0.797MSTPSERM
Evaluation Results
| Method | Links | |
|---|---|---|
| ERM2026.04 | 0.797 | |
| SHORTCUT GUARDRAIL2026.04 | 0.381 | |
| JTT2026.04 | 0.267 | |
| DFR2026.04 | 0.212 | |
| NFL2026.04 | 0.143 |
Loading the SOTA2 catalog…