Mathematical Reasoning on AIME (Exact Match Accuracy)
12.31Exact Match AccuracyExplorer-only, successful seeds (diagnostic)
Evaluation Results
| Method | Links | |
|---|---|---|
| Explorer-only, successful seeds (diagnostic)Backbone=Qwen3-8B-Base2026.05 | 12.31 | |
| Entropy reweightingBackbone=Qwen3-8B-Base, Layer=Mid (Layer 20)2026.05 | 9.75 | |
| Entropy reweightingBackbone=Qwen3-8B-Base, Layer=Deep (Layer 31)2026.05 | 9.14 | |
| Random-20%Backbone=Qwen3-8B-Base2026.05 | 8.12 | |
| Entropy reweightingBackbone=Qwen3-8B-Base, Layer=Shallow (Layer 8)2026.05 | 7.37 | |
| Full-token DAPOBackbone=Qwen3-8B-Base2026.05 | 5.83 | |
| Anchor-onlyBackbone=Qwen3-8B-Base2026.05 | 5.6 | |
| Explorer-only, all seedsBackbone=Qwen3-8B-Base2026.05 | 4.62 | |
| High Prediction Entropy-onlyBackbone=Qwen3-8B-Base2026.05 | 1.84 |