Mathematical Reasoning on Average (GSM8k, GSM-sym, MATH, NuminaMath, AIME)
71.41AccuracySA-GRPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SA-GRPOBase Model=Qwen3-4B, alpha=0.22025.09 | 71.41 | 894 | 1.564 | |
| GRPOBase Model=Qwen3-4B2025.09 | 69.07 | 1,030.6 | 1.165 | |
| GRPO-O1Base Model=Qwen3-4B, alpha=0.052025.09 | 67.75 | 608.6 | 1.592 | |
| GRPO-ERBase Model=Qwen3-4B, alpha=0.052025.09 | 67.38 | 579 | 1.596 | |
| PPOBase Model=Qwen3-4B2025.09 | 67.33 | 1,333 | 0.606 | |
| SA-PPOBase Model=Qwen3-4B2025.09 | 66.32 | 758.6 | 1.26 | |
| SA-GRPOBase Model=Qwen3-1.7B, alpha=0.22025.09 | 54.13 | 602 | 1.795 | |
| GRPOBase Model=Qwen3-1.7B2025.09 | 53.37 | 762.8 | 1.509 | |
| Qwen3-4BBase Model=Qwen3-4B2025.09 | 53.03 | 765 | 0 | |
| GRPO-O1Base Model=Qwen3-1.7B, alpha=0.052025.09 | 52.64 | 572.2 | 1.652 | |
| SA-PPOBase Model=Qwen3-1.7B2025.09 | 52.49 | 623.6 | 1.572 | |
| GRPO-ERBase Model=Qwen3-1.7B, alpha=0.052025.09 | 52.11 | 666.6 | 1.474 | |
| PPOBase Model=Qwen3-1.7B2025.09 | 48.95 | 930.4 | 0.773 | |
| SA-GRPOBase Model=Phi-3.5-mini, alpha=0.22025.09 | 46.19 | 368.4 | 2.137 | |
| GRPOBase Model=Phi-3.5-mini2025.09 | 45.91 | 448.2 | 2.003 | |
| SA-PPOBase Model=Phi-3.5-mini2025.09 | 45.42 | 393.8 | 1.997 | |
| Qwen3-1.7BBase Model=Qwen3-1.7B2025.09 | 41.48 | 825.4 | 0 | |
| GRPO-ERBase Model=Phi-3.5-mini, alpha=0.052025.09 | 40.84 | 374.8 | 1.368 | |
| GRPO-O1Base Model=Phi-3.5-mini, alpha=0.052025.09 | 40.62 | 392.8 | 1.316 | |
| PPOBase Model=Phi-3.5-mini2025.09 | 40.24 | 799 | 0.784 | |
| Phi-3.5-miniBase Model=Phi-3.5-mini2025.09 | 35.15 | 850.4 | 0 | |
| SA-GRPOBase Model=Gemma3-1B, alpha=0.22025.09 | 32.24 | 1,017.4 | 2.218 | |
| GRPOBase Model=Gemma3-1B2025.09 | 31.78 | 1,866 | 1.343 | |
| GRPO-ERBase Model=Gemma3-1B, alpha=0.052025.09 | 31.29 | 1,236.6 | 1.808 | |
| GRPO-O1Base Model=Gemma3-1B, alpha=0.052025.09 | 31.23 | 1,529.6 | 1.528 | |
| PPOBase Model=Gemma3-1B2025.09 | 29.69 | 1,572.6 | 1.146 | |
| SA-PPOBase Model=Gemma3-1B2025.09 | 29.1 | 1,170 | 1.383 | |
| Gemma3-1BBase Model=Gemma3-1B2025.09 | 22.56 | 1,097 | 0 |