Source and dataset notices
Author-controlled source contributions use the included Apache-2.0 license. The model-weight license remains unassigned; this source license does not make an additional weight-license grant. Installed dependencies retain their own terms: MLX (MIT), NumPy (BSD-3-Clause), Tokenizers (Apache-2.0).
Pretraining corpus declarations, revisions and filters are in corpus.json:
| Source | Recorded source declaration |
|---|---|
| FineWeb-Edu through SmolLM corpus | ODC-BY; underlying content notices retained |
| DCLM-Edu | CC-BY-4.0 |
| English FineWiki | CC-BY-SA-4.0 / GFDL |
| Cosmopedia v2 through SmolLM corpus | ODC-BY; synthetic-data provenance retained |
| Python text mirror | Mirror metadata ODC-BY; original code remains subject to upstream repository licenses and The Stack v2 terms |
The pinned Python mirror card was unavailable during the preceding report audit; its declaration is taken from the recorded local corpus specification, not a newly completed audit of underlying content. Dataset declarations are not blanket redistribution permission for all contained source documents or model weights. This package redistributes no raw pretraining documents. It includes own qualitative prompts/generations and aggregate historical validation results. No SQuAD, Dolly or SNIPS SFT is included in this base checkpoint.