---
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|