Source attribution
This release preserves source identifiers, revisions, URLs, row IDs and record-level licenses in the Parquet files. Detailed source cards are available under metadata/source_cards/.
Public CPT sources
CommitPackFT
- Source:
bigcode/commitpackft - Upstream revision:
fc56fe33c030c6daa414c2b112c932b8eed085e6 - Dataset wrapper: MIT
- Record licensing: per-repository license stored in each row
- Attribution data: repository, path/commit provenance and source row ID
Only single-repository records with allowlisted licenses were selected.
Russian Wikipedia
- Source:
wikimedia/wikipedia, Russian configuration - License: CC-BY-SA-3.0/GFDL
- Required attribution: article URL/title and Wikipedia contributors
Public SFT sources
Aya
- Source:
CohereForAI/aya_dataset - License: Apache-2.0
Databricks Dolly 15k
- Source:
databricks/databricks-dolly-15k - License: CC-BY-SA-3.0
OpenAssistant OASST1
- Source:
OpenAssistant/oasst1 - License: Apache-2.0
- Attribution: OpenAssistant project and Apache-2.0 notice
OpenAssistant OASST2
- Source:
OpenAssistant/oasst2 - License: Apache-2.0
SmolTalk selected subsets
- Source:
HuggingFaceTB/smoltalk - Included subsets: native Apache-2.0 subsets selected by the curation pipeline
WikiOmnia
- Source:
RussianNLP/wikiomnia - Wrapper: Apache-2.0
- Upstream content: CC-BY-SA-3.0 Wikipedia material
- Attribution: WikiOmnia authors/paper, Wikipedia article metadata and CC-BY-SA notice
Project-authored identity examples
- Source ID:
wex_identity - Original records authored for the WexCode-RU curriculum
- Released for use and redistribution as part of RuEn Code Curriculum
CommitPackFT is also used in the SFT configuration. Its record-level repository licenses and provenance remain attached to every example.