Wisp Coder 110M — NOTICE

This is the attribution notice for the Wisp Coder 110M release artifact
(weights, configuration, and reference code), which is licensed under the
Apache License, Version 2.0. See the LICENSE file for the full license text.

Training data
-------------

The model was trained on the following sources. These remain subject to their
original terms and licenses; this release does not relicense them, and it makes
no per-row license or attribution guarantee.

- bigcode/starcoderdata (Hugging Face). Declared `license: other` on the Hub;
  92 percent of the configured training-token target source weight. The
  original repositories and any applicable BigCode or per-repository terms,
  and relevant attribution clauses, continue to apply.
- HuggingFaceFW/fineweb-edu (Hugging Face). ODC-By 1.0; 8 percent of the
  configured training-token target source weight. Remains subject to Common
  Crawl terms.

Run 1 retained no row-level source manifest, so per-file licensing or
attribution for individual training rows cannot be established from the
release. Full disclosure is in the model card, DATA.md, and
config/training_data_receipt.json.

Dependencies
------------

- mlx / mlx-lm (Apple) — training and reference runtime. Subject to its own
  license; included by reference, not copied into this package.
- tokenizers (Hugging Face) — tokenizer runtime. Subject to its own license;
  included by reference.

This NOTICE is not a license. The Apache 2.0 terms describe the released Wisp
artifact and do not override training-source terms or licenses applicable to
generated code.