Mathematical Reasoning on SVAMP (FRS, pass@1 accuracy)
94.3Pass@1 AccuracyMT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MTTeacher Model=Qwen2-Math-7B-Instruct2026.04 | 94.3 | — | |
| VCRDTeacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 85.9 | — | |
| DistilLLMTeacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 85.8 | — | |
| ABKDTeacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 85.5 | — | |
| DistillLM-2Teacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 85 | — | |
| GKDTeacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 84.7 | — | |
| KDTeacher Model=Qwen2-Math-7B-Instruct, Student Model=Qwen2-Math-1.5B2026.04 | 84.3 | — | |
| AdapShotBackbone=Qwen2.5-7B2026.05 | 79 | — | |
| DBSABackbone=Qwen2.5-7B2026.05 | 73.7 | — | |
| Many-shot (512)Backbone=Qwen2.5-7B2026.05 | 45.6 | — | |
| MSStudent Model=Qwen2-Math-1.5B2026.04 | 21.7 | — | |
| Many-shot (256)Backbone=Qwen2.5-7B2026.05 | 14.6 | — | |
| Zero-shotBackbone=Qwen2.5-7B2026.05 | 13.6 | — | |
| Few-shot (8)Backbone=Qwen2.5-7B2026.05 | 10.7 | — | |
| DS-R1-1.5BFiltering rate=10%, Decoding strategy=greedy2026.04 | — | 96.9 | |
| DS-R1-7BFiltering rate=10%, Decoding strategy=greedy2026.04 | — | 99 | |
| Gemma-7BFiltering rate=10%, Decoding strategy=greedy2026.04 | — | 31.1 | |
| LLaMA-3.1-8BFiltering rate=10%, Decoding strategy=greedy2026.04 | — | 80 | |
| Phi-4Filtering rate=10%, Decoding strategy=greedy2026.04 | — | 75.5 | |
| Phi-4-Reas.Filtering rate=10%, Decoding strategy=greedy2026.04 | — | 74.8 | |
| Qwen2.5-7BFiltering rate=10%, Decoding strategy=greedy2026.04 | — | 64.9 | |
| Qwen2.5-MathFiltering rate=10%, Decoding strategy=greedy2026.04 | — | 74.5 | |
| Qwen3-4BFiltering rate=10%, Decoding strategy=greedy2026.04 | — | 91 |