Mathematical Reasoning on GSM8K (Acc, Delta_Up, Delta_Down)
94.6AccuracyAgentic-Verifier-Qwen3-4B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Agentic-Verifier-Qwen3-4BRefinement Turn=12026.04 | 94.6 | 14.5 | 0.5 | |
| Agentic-Verifier-Qwen3-4BRefinement Turn=22026.04 | 94.1 | 0.4 | 0.9 | |
| Agentic-Verifier-Qwen3-4BRefinement Turn=32026.04 | 94.1 | 1 | 1 | |
| DS-Distill-14BRefinement Turn=32026.04 | 93.4 | 1.8 | 1.1 | |
| DS-Distill-14BRefinement Turn=22026.04 | 92.7 | 2.4 | 1.7 | |
| Qwen3-4B-ThinkRefinement Turn=22026.04 | 92.2 | 2.4 | 1.1 | |
| DS-Distill-14BRefinement Turn=12026.04 | 92 | 12.1 | 0.8 | |
| Qwen3-4B-ThinkRefinement Turn=32026.04 | 92 | 1.2 | 1.4 | |
| Qwen3-4B-ThinkRefinement Turn=12026.04 | 90.9 | 11.1 | 0.9 | |
| Mistral-Small-24B-InstructRefinement Turn=32026.04 | 90.6 | 2.65 | 2.05 | |
| Mistral-Small-24B-InstructRefinement Turn=22026.04 | 90 | 2.7 | 1.1 | |
| Mistral-Small-24B-InstructRefinement Turn=12026.04 | 88.4 | 8.5 | 0.8 | |
| Qwen2.5-7B-InstructRefinement Turn=32026.04 | 87.5 | 2.96 | 2.8 | |
| Qwen2.5-7B-InstructRefinement Turn=22026.04 | 87.4 | 3.3 | 1.9 | |
| Qwen2.5-7B-InstructRefinement Turn=12026.04 | 85.9 | 5.9 | 0.6 | |
| Llama3.1-8B-InstructRefinement Turn=12026.04 | 84.1 | 4.6 | 1.1 | |
| Llama3.1-8B-InstructRefinement Turn=22026.04 | 82.8 | 3.3 | 4.6 | |
| Llama3.1-8B-InstructRefinement Turn=32026.04 | 82.8 | 3.3 | 3.3 |