Mathematical Reasoning on GSM8K (maj@16, pass@16)
97.12maj@16 AccuracyPPC
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PPCBackbone=Qwen2.5-Math-7B2026.05 | 97.12 | 98.56 | |
| Prompt-OnlyBackbone=Qwen2.5-Math-7B2026.05 | 97.04 | 98.18 | |
| BaseBackbone=Qwen2.5-Math-7B2026.05 | 96.97 | 98.1 | |
| Plan-TuningBackbone=Qwen2.5-Math-7B2026.05 | 96.82 | 98.33 | |
| GRPOBackbone=Qwen2.5-Math-7B2026.05 | 96.36 | 98.4 | |
| PTA-GRPOBackbone=Qwen3-4B2026.05 | 95.3 | 97.57 | |
| PPCBackbone=Qwen2.5-7B2026.05 | 95.22 | 98.48 | |
| PPCBackbone=Qwen3-4B2026.05 | 95.15 | 98.03 | |
| PTA-GRPOBackbone=Qwen2.5-Math-7B2026.05 | 95.15 | 98.03 | |
| Plan-TuningBackbone=Qwen3-4B2026.05 | 95 | 97.12 | |
| GRPOBackbone=Qwen3-4B2026.05 | 94.92 | 97.04 | |
| BaseBackbone=Qwen3-4B2026.05 | 94.84 | 96.89 | |
| Prompt-OnlyBackbone=Qwen3-4B2026.05 | 94.62 | 97.27 | |
| PTA-GRPOBackbone=Qwen2.5-7B2026.05 | 94.62 | 97.27 | |
| Prompt-OnlyBackbone=Qwen2.5-7B2026.05 | 94.47 | 97.12 | |
| GRPOBackbone=Qwen2.5-7B2026.05 | 94.31 | 97.57 | |
| Plan-TuningBackbone=Qwen2.5-7B2026.05 | 94.09 | 97.19 | |
| BaseBackbone=Qwen2.5-7B2026.05 | 93.71 | 97.04 | |
| PPCBackbone=Llama3.1-8B2026.05 | 92.04 | 98.03 | |
| Plan-TuningBackbone=Llama3.1-8B2026.05 | 90.83 | 96.13 | |
| PTA-GRPOBackbone=Llama3.1-8B2026.05 | 90.75 | 96.59 | |
| GRPOBackbone=Llama3.1-8B2026.05 | 90.59 | 96.36 | |
| Prompt-OnlyBackbone=Llama3.1-8B2026.05 | 90.37 | 96.47 | |
| BaseBackbone=Llama3.1-8B2026.05 | 89.76 | 95.98 |