Math Reasoning on AMC 2023 (avg@16, pass@16)
87.5Avg@16KPO-clipped
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| KPO-clippedClipping strategy=clipped, Number of candidate solutions=162026.02 | 87.5 | 95 | |
| KPO-unclippedClipping strategy=unclipped, Number of candidate solutions=162026.02 | 80 | 95 | |
| IRDSModel=QWEN3-4B, Data Budget=20%2026.05 | 79.5 | — | |
| GMPONumber of candidate solutions=162026.02 | 76.56 | 87.5 | |
| GSPONumber of candidate solutions=162026.02 | 75.46 | 95 | |
| GRPONumber of candidate solutions=162026.02 | 73.43 | 92.5 | |
| PPL-TOPModel=QWEN3-4B, Data Budget=20%2026.05 | 72.4 | — | |
| DEPOModel=QWEN3-4B, Data Budget=20%2026.05 | 69 | — | |
| IFDModel=QWEN3-4B, Data Budget=20%2026.05 | 68.9 | — | |
| LIMRModel=QWEN3-4B, Data Budget=20%2026.05 | 68.6 | — | |
| PPL-MIDDLEModel=QWEN3-4B, Data Budget=20%2026.05 | 67.7 | — | |
| RANDOMModel=QWEN3-4B, Data Budget=20%2026.05 | 62.4 | — | |
| IRDSModel=QWEN3-1.7B, Data Budget=20%2026.05 | 55.8 | — | |
| TOKEN-LENGTHModel=QWEN3-4B, Data Budget=20%2026.05 | 55.2 | — | |
| RANDOMModel=QWEN3-1.7B, Data Budget=20%2026.05 | 48.9 | — | |
| DEPOModel=QWEN3-1.7B, Data Budget=20%2026.05 | 47.6 | — | |
| LIMRModel=QWEN3-1.7B, Data Budget=20%2026.05 | 45.1 | — | |
| PPL-TOPModel=QWEN3-1.7B, Data Budget=20%2026.05 | 45 | — | |
| IFDModel=QWEN3-1.7B, Data Budget=20%2026.05 | 43.2 | — | |
| PPL-MIDDLEModel=QWEN3-1.7B, Data Budget=20%2026.05 | 42.4 | — | |
| TOKEN-LENGTHModel=QWEN3-1.7B, Data Budget=20%2026.05 | 37.5 | — | |
| TOKEN-LENGTHModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=20%2026.05 | 23.1 | — | |
| RANDOMModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=20%2026.05 | 23 | — | |
| DEPOModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=20%2026.05 | 22.8 | — | |
| LIMRModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=20%2026.05 | 22.3 | — | |
| IRDSModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=20%2026.05 | 21.7 | — | |
| PPL-TOPModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=20%2026.05 | 20.7 | — | |
| PPL-MIDDLEModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=20%2026.05 | 20.4 | — | |
| IFDModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=20%2026.05 | 19.8 | — |