Mathematical Reasoning on GSM8K (avg@3 Accuracy)
95.9avg@3 AccuracyGemma-3-27B-IT
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemma-3-27B-ITmax_completion_tokens=8K2026.01 | 95.9 | |
| Qwen3-8B-Base + PASC-GRPOmax_completion_tokens=8K2026.01 | 95.37 | |
| Gemma-3-12B-ITmax_completion_tokens=8K2026.01 | 94.4 | |
| Qwen3-4B-Base + PASC-GRPOmax_completion_tokens=8K2026.01 | 90.45 | |
| Qwen3-8B-Base + GRP-SFTmax_completion_tokens=8K2026.01 | 87.72 | |
| LLaMA-3.1-8B-Instructmax_completion_tokens=8K2026.01 | 85 | |
| Qwen3-4B-Base + GRP-SFTmax_completion_tokens=8K2026.01 | 82.3 | |
| Qwen3-8B-Basemax_completion_tokens=8K2026.01 | 73.01 | |
| Qwen3-4B-Basemax_completion_tokens=8K2026.01 | 70.2 |