Logical Reasoning on BBH (Accuracy, Loss)
52.5Exact Match Accuracy (BBH Logical Reasoning)Dream-7B-Base + DyStruct
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Dream-7B-Base + DyStructBackbone=Dream-7B, Decoding method=DyStruct, Few-shot examples=3, Unmasking iterations=256, Generation limit=2562026.05 | 52.5 | — | — | |
| Dream-7B-BaseBackbone=Dream-7B, Decoding method=Base, Few-shot examples=3, Unmasking iterations=256, Generation limit=2562026.05 | 51.7 | — | — | |
| LLaDA-8B-Base + DyStructBackbone=LLaDA-8B, Decoding method=DyStruct, Few-shot examples=3, Unmasking iterations=256, Generation limit=2562026.05 | 49.3 | — | — | |
| LLaDA-8B-BaseBackbone=LLaDA-8B, Decoding method=Base, Few-shot examples=3, Unmasking iterations=256, Generation limit=2562026.05 | 44.9 | — | — | |
| Dream-7B-Base + DAEDALBackbone=Dream-7B, Decoding method=DAEDAL, Few-shot examples=3, Unmasking iterations=256, Generation limit=2562026.05 | 44.8 | — | — | |
| LLaDA-8B-Base + DAEDALBackbone=LLaDA-8B, Decoding method=DAEDAL, Few-shot examples=3, Unmasking iterations=256, Generation limit=2562026.05 | 43.7 | — | — | |
| AdamModel scale=1B2026.04 | — | 24.8 | 1.7 | |
| AdamModel scale=3B2026.04 | — | 47.4 | 1.529 | |
| NexusModel scale=1B2026.04 | — | 23 | 1.689 | |
| NexusModel scale=3B2026.04 | — | 44.4 | 1.54 |