metadata
license: apache-2.0
Chinese wikipedia (sourced from here) cleaned by the following steps:
- All text in Traditional Chinese is converted into Simplified Chinese using zhconv.
- Duplicates are removed.
- Low-quality text are heuristically removed based on proportion of special characters.
- Qwen2.5-32B-Instruct is used to generate language quality annotation (on a scale of 1-5) for 398K Chinese samples and 250K English samples. An XLM-RoBERT-large classifier is trained with regression on these annotations. Any document receiving a score of 1 or 2 from the classifier is removed.
Data Statistics
Number of samples: 2,021,184.
Size of parquet files: 3G.