back-prop/

RLVR-datasets

Dataset files for the VO/RLVR experiment configured in vo/tmp/run_vo.sh.

train/math__combined_54.4k.parquet: training set used by TRAIN_FILE; 54,404 rows. validation/math__aime_repeated_32x_960.parquet: AIME validation set used by VAL_AIME_FILE; 960 rows. validation/math__math_500.parquet: MATH-500 validation set used by VAL_MATH500_FILE; 500 rows.

The parquet files preserve the local schemas used by the verl/RLVR training script, including prompt… See the full description on the dataset page: https://huggingface.co/datasets/back-prop/RLVR-datasets.