Cronus 5B training package
5,000,001,536 training tokens in 2,441,407 rows of 2048. uint16 tokens, uint8 source IDs.
Ordered prefix of Slice2 at revision a9e50f939e7c464de938bcf0079ed4fd38ca97a3. 1B is contained in 5B, which is contained in Slice2 10B; these are not independent datasets. The original Slice2 held-out probe is included unchanged and is not counted as training tokens. Original cleaning and provenance: source_manifest.json. No new source sampling or data cleaning. Scope remains slice.