Pre-processed parquet files used by the RLCSD training code.
Each subdirectory is one dataset; the file inside is the consumed split.
deepmath_filtered_level5_7/train.parquet
train
Qwen3-1.7B math (DeepMath levels 5–7)
deepmath_filtered_level6_8/train.parquet
train
Qwen3-4B math (DeepMath levels 6–8)
deepmath_filtered_level7_10/train.parquet
train
Qwen3-8B + Olmo3-7B-Think math (7–10)… See the full description on the dataset page:
https://huggingface.co/datasets/Leyiii/RLCSD.