Mathematical Reasoning on MATH 500 (average@10)
48.5Average@10 ScoreQwen2.5-3B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-3BTraining Setting=RL fine-tuned with REVERSE-MATH-augmented samples2026.05 | 48.5 | |
| Qwen2.5-3BTraining Setting=RL fine-tuned with duplicated GSM8K samples2026.05 | 46.2 | |
| Qwen2.5-3BTraining Setting=Untuned2026.05 | 45.6 | |
| Llama3.1-8BTraining Setting=RL fine-tuned with duplicated GSM8K samples2026.05 | 32.8 | |
| Llama3.2-3BTraining Setting=RL fine-tuned with REVERSE-MATH-augmented samples2026.05 | 31.4 | |
| Llama3.1-8BTraining Setting=RL fine-tuned with REVERSE-MATH-augmented samples2026.05 | 30.9 | |
| Llama3.2-3BTraining Setting=Untuned2026.05 | 30.7 | |
| Llama3.2-3BTraining Setting=RL fine-tuned with duplicated GSM8K samples2026.05 | 30.1 | |
| Llama3.1-8BTraining Setting=Untuned2026.05 | 29.7 | |
| Mistral-7BTraining Setting=RL fine-tuned with REVERSE-MATH-augmented samples2026.05 | 15 | |
| Mistral-7BTraining Setting=Untuned2026.05 | 14.9 | |
| Mistral-7BTraining Setting=RL fine-tuned with duplicated GSM8K samples2026.05 | 13 |