Dataset Viewer (First 5GB)
Auto-converted to Parquet
The dataset viewer is not available for this split.
Job manager crashed while running this job (missing heartbeats).
Error code:   JobManagerCrashedError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

A model-filtered version of C4, including two stages:

Removing Advertisements

A RoBERTa-large classifier is trained on 550K binary annotations by Qwen2.5-32B-Instruct. The classifier is then used to filter out advertisements in the whole corpus (see here for the removed advertisements).

Removing documents with low-quality language

Qwen2.5-32B-Instruct is used again to generate language quality annotation (on a scale of 1-5) for 250K samples. A RoBERT-large classifier is trained with regression on these annotations. Any document receiving a score of 1 or 2 from the classifier is removed.

Dataset Statistics

Number of samples: 241,791,494.

Size of downloaded parquet files: 397G.

Downloads last month
610