Mathematical Reasoning on GSM8K (average@10)
68.2Average Score @10Qwen2.5-3B
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen2.5-3BTraining Setting=RL fine-tuned with duplicated GSM8K samples2026.05 | 68.2 | |
| Qwen2.5-3BTraining Setting=RL fine-tuned with REVERSE-MATH-augmented samples2026.05 | 67.2 | |
| Qwen2.5-3BTraining Setting=Untuned2026.05 | 64.9 | |
| Llama3.1-8BTraining Setting=Untuned2026.05 | 52.1 | |
| Llama3.1-8BTraining Setting=RL fine-tuned with REVERSE-MATH-augmented samples2026.05 | 51.8 | |
| Llama3.2-3BTraining Setting=RL fine-tuned with duplicated GSM8K samples2026.05 | 50.7 | |
| Llama3.2-3BTraining Setting=RL fine-tuned with REVERSE-MATH-augmented samples2026.05 | 50.1 | |
| Llama3.1-8BTraining Setting=RL fine-tuned with duplicated GSM8K samples2026.05 | 49.5 | |
| Llama3.2-3BTraining Setting=Untuned2026.05 | 48.1 | |
| Mistral-7BTraining Setting=RL fine-tuned with duplicated GSM8K samples2026.05 | 29.7 | |
| Mistral-7BTraining Setting=RL fine-tuned with REVERSE-MATH-augmented samples2026.05 | 29.5 | |
| Mistral-7BTraining Setting=Untuned2026.05 | 28.2 |