metadata
license: mit
Consistent Paths Lead to Truth: Self-Rewarding Reinforcement Learning for LLM Reasoning
The file test.jsonl is used to monitor the model performance as training proceeds.
| Level | Source |
|---|---|
| 1 | GSM8K |
| 2 | MATH-500 |
| 3 | AMC-23 |
| 4 | AIME2024 |
| 5 | AIME2025 |