The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
Nebula public raw corpus
This dataset contains a bounded sample of public-source UTF-8 documents packed as little-endian uint16 byte token IDs. BOS=256 and EOS=257; IDs 0-255 encode bytes. The index preserves whole-document boundaries, source attribution and train/validation assignment. No Nebula model checkpoint, private account-chat export, review feed, token or API configuration is uploaded. This is a research corpus, not a conversational model or a claim of verified factual quality. Original public web documents may themselves contain personal information; the filters are not a full PII audit.
Licenses and attribution
Each component retains its upstream license and terms; this collection does not relicense them.
Read the pinned upstream cards retained under attribution/ and original source pages.
FineWeb-Edu also retains the original source URLs and Common Crawl attribution in the index.
ODC-By terms: https://opendatacommons.org/licenses/by/1-0/
TinyStories CDLA terms: https://cdla.dev/sharing-1-0/
- HuggingFaceFW/fineweb-edu, subset
sample-10BT, licenseodc-by. - HuggingFaceTB/smollm-corpus, subset
cosmopedia-v2, licenseodc-by. - roneneldan/TinyStories, subset
default, licensecdla-sharing-1.0.
Snapshot format
public-state.json points to a content-addressed snapshot. Each file or part has a SHA256 and byte count.
Concatenate .partNNN.bin files in manifest order to restore a sharded file.
manifest.json reports whether the requested corpus preparation is complete; partial snapshots can resume.
documents.sqlite.bin is a fresh SQLite database containing only the validated corpus index and preparation metadata.
No previous repository history is imported. No source files are deleted by this transfer.
- Downloads last month
- -