metadata
license: odc-by
language:
- ja
- en
This pre-training dataset was created for shisa-base-7b-v1.
It is primarily composed of a DSIR sampling of MADLAD-400 JA/EN tokens in a 90%/10% ratio.
license: odc-by
language:
- ja
- en
This pre-training dataset was created for shisa-base-7b-v1.
It is primarily composed of a DSIR sampling of MADLAD-400 JA/EN tokens in a 90%/10% ratio.