Multiple-Choice Question Answering on ARC Challenge (Accuracy and Improvement)
86.1AccuracyReflective Masking (Ours)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Reflective Masking (Ours)Base Model=LLaDA2026.06 | 86.1 | 4.8 | |
| Vanilla SFTBase Model=LLaDA2026.06 | 81.3 | — | |
| LLaDA2026.06 | 73.7 | — |