File size: 425 Bytes
a7d9d3c | 1 2 3 4 5 6 7 | # SCOPE-RL reproduction data
Prepared from the SCOPE-RL paper repo (github.com/tokencraft-lab/SCOPE-RL, arXiv 2607.11506) by `scope_rl/prep_data.py` in github.com/alphaXiv/paper-implementations.
- `dapo_math_2400.jsonl` - DAPO-Math 2400 training subset: chat prompt + rule-verifiable ground truth (sub-question decompositions stripped)
- `benchmark.jsonl` - paper eval benchmark: AIME24/25 (x8 rows), MATH500, GPQA-diamond
|