Mathematical Reasoning on Countdown-34 (held-out)
81.26AccuracyDS (Oracle)
Evaluation Results
| Method | Links | |
|---|---|---|
| DS (Oracle)Backbone=Qwen2.5-7B, Rollouts=1006k, Finetuned=true2026.03 | 81.26 | |
| DPSBackbone=Qwen2.5-7B, Rollouts=246k, Finetuned=true2026.03 | 81.15 | |
| HRBackbone=Qwen2.5-7B, Rollouts=246k, Finetuned=true2026.03 | 78.15 | |
| USBackbone=Qwen2.5-7B, Rollouts=246k, Finetuned=true2026.03 | 77.84 | |
| DS (Oracle)Backbone=Qwen2.5-3B, Rollouts=1141k, Finetuned=true2026.03 | 74.95 | |
| DPSBackbone=Qwen2.5-3B, Rollouts=246k, Finetuned=true2026.03 | 74.27 | |
| HRBackbone=Qwen2.5-3B, Rollouts=246k, Finetuned=true2026.03 | 70.19 | |
| USBackbone=Qwen2.5-3B, Rollouts=246k, Finetuned=true2026.03 | 69.87 |