Datasetpublic

Publishing/unclutching-corpus-v2

unclutching-corpus-v2 Mixed continued-pretraining corpus on Unclutching combining synthesized expository passages with deduplicated excerpts from the Kailasa wiki book corpus. Format JSONL. Every row has text and source_dataset; rows from the wiki also carry origin metadata. {"text": "...", "source_dataset": "synth-cpt"} {"text": "...", "source_dataset": "kailasa-wiki", "corpus": "book", "source": "...", "id": "..."} Composition… See the full description on the dataset page: https://huggingface.co/datasets/Publishing/unclutching-corpus-v2.

sourceHugging Faceotherupdated 5mo agoView on Hugging Face

0likes15downloads

Dataset Card

unclutching-corpus-v2

Mixed continued-pretraining corpus on Unclutching combining synthesized expository passages with deduplicated excerpts from the Kailasa wiki book corpus.

Format

JSONL. Every row has text and source_dataset; rows from the wiki also carry origin metadata.

json

{"text": "...", "source_dataset": "synth-cpt"}
{"text": "...", "source_dataset": "kailasa-wiki", "corpus": "book", "source": "...", "id": "..."}

Composition

source_dataset	rows
synth-cpt	194
kailasa-wiki	3,766
total	3,960

Notes

—synth-cpt rows: synthesized long-form English passages from synth-cpt-cli.
—kailasa-wiki rows: deduplicated excerpts from the Kailasa wiki book exports (multilingual; field originally named excerpt, mapped to text).