---
configs:
- config_name: default
  data_files: 'data/*.parquet'
---
# Beetle-Data/zh-28B

Pretokenized chunks (`input_ids` = 513-token packed sequences,
no cross-document bleeding). Sharded incrementally; the marker
`data/_finalized.json` is committed once all parts are uploaded.