Mathematical Reasoning on MathQA (Exact Match)
52.4Exact MatchAdapShot
Evaluation results
| Method | Links | |
|---|---|---|
| AdapShotBackbone=Qwen2.5-7B2026.05 | 52.4 | |
| Many-shot (256)Backbone=Qwen2.5-7B2026.05 | 48.3 | |
| TILRBackbone=GPT-2, Average generation tokens per inference=92026.06 | 42.1 | |
| RefinementBackbone=GPT-2, Average generation tokens per inference=92026.06 | 40.1 | |
| CoconutBackbone=GPT-2, Average generation tokens per inference=92026.06 | 38.5 | |
| AdaAnchorBackbone=GPT-2, Average generation tokens per inference=92026.06 | 37.5 | |
| Few-shot (8)Backbone=Qwen2.5-7B2026.05 | 35 | |
| CoTBackbone=GPT-2, Average generation tokens per inference=462026.06 | 34.2 | |
| DBSABackbone=Qwen2.5-7B2026.05 | 32.3 | |
| No-CoTBackbone=GPT-2, Average generation tokens per inference=32026.06 | 24.5 | |
| Many-shot (512)Backbone=Qwen2.5-7B2026.05 | 23.2 | |
| Zero-shotBackbone=Qwen2.5-7B2026.05 | 1.2 |