Math Reasoning on MATH 500 (multi@5 accuracy)
58.2Multi@5 AccuracyDRIFT-5turn
Evaluation Results
| Method | Links | |
|---|---|---|
| DRIFT-5turnBackbone=Qwen2.5-3B-Instruct, Protocol=Ours, Max turns=52026.05 | 58.2 | |
| UFO-5turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=52026.05 | 56.4 | |
| SCoRe-2turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=22026.05 | 55.8 | |
| SFT-5turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=52026.05 | 53.6 | |
| PPOBackbone=Qwen2.5-3B-Instruct, Protocol=Single-turn baselines2026.05 | 51.2 | |
| SFTBackbone=Qwen2.5-3B-Instruct, Protocol=Single-turn baselines2026.05 | 50.8 | |
| STaR-2turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=22026.05 | 50.4 | |
| DRIFT-5turnBackbone=Llama3.1-8B-Instruct, Protocol=Ours, Max turns=52026.05 | 48.2 | |
| UFO-5turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=52026.05 | 46.4 | |
| SCoRe-2turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=22026.05 | 45 | |
| SFT-5turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=52026.05 | 44 | |
| STaR-2turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=22026.05 | 43.2 | |
| SFTBackbone=Llama3.1-8B-Instruct, Protocol=Single-turn baselines2026.05 | 42 | |
| PPOBackbone=Llama3.1-8B-Instruct, Protocol=Single-turn baselines2026.05 | 41.6 | |
| BaseBackbone=Qwen2.5-3B-Instruct, Protocol=Single-turn baselines2026.05 | 40.2 | |
| BaseBackbone=Llama3.1-8B-Instruct, Protocol=Single-turn baselines2026.05 | 38.2 |