# Third-party data and attribution

This is a practical attribution index, not legal advice. Review the upstream terms for the
intended distribution and deployment. Exact repository revisions and realized counts are
recorded in `manifest.json`.

| Partition | Upstream | Selected terms / policy |
|---|---|---|
| `open_code_instruct` | `nvidia/OpenCodeInstruct` | CC BY 4.0; perfect executed-test rows only |
| `ling_coder` | `inclusionAI/Ling-Coder-SFT` | Apache 2.0 |
| `code_fim`, `czech_code` | Derived from the two code sources | Parent terms and lineage retained |
| `open_code_reasoning` | `nvidia/OpenCodeReasoning-2` + `codeparrot/apps` | Attribution retained; APPS train prompts are MIT |
| `open_swe` | `nvidia/Open-SWE-Traces` v1.0 | CC BY 4.0 plus per-row permissive repository license |
| `terminal` | `nvidia/Nemotron-Terminal-Corpus` | CC BY 4.0 |
| `chat` | `nvidia/Nemotron-SFT-Instruction-Following-Chat-v3` | CC BY 4.0 and ODC-By |
| `eu_czech` | `openeurollm/EU-Instruct-Synthetic`, `cs` | Apache 2.0 |
| `muri_czech` | `akoksal/muri-it-language-split`, `ces` | Apache 2.0; original subdataset retained per row |
| `math` | `nvidia/Nemotron-SFT-Math-v4` | Rows explicitly marked CC BY 4.0 only |
| `science` | `nvidia/OpenScienceReasoning-2` | CC BY 4.0 |
| `agentic` | `nvidia/Nemotron-SFT-Agentic-v2` | CC BY 4.0; card also identifies Apache 2.0/MIT inputs |
| `structured` | `nvidia/Nemotron-Instruction-Following-Chat-v1` | CC BY 4.0 rows only |
| `safety` | `nvidia/Nemotron-SFT-Safety-v1` | CC BY 4.0 rows only |
| `identity` | Locally generated | LuxAI-generated |

Keep this file and `manifest.json` with redistributed copies. Upstream dataset URLs are
`https://huggingface.co/datasets/<repo>` for the repository names above.