Engineering, production, governance, cost, communication · insight-0280
Check model and dataset licences before a prototype becomes a product
- Evidence type
- documented fact
- Difficulty
- easy
- Roles
- Applied AI, ML engineer
- Industries
- cross-industry
Summary: situation, action, reason, boundary
Situation
A prototype uses an open-weights model and a public dataset, and the business wants to launch it commercially.
Action
Record the licence, version and source of every model, dataset and checkpoint in the registry; review non-standard or 'other' licences with legal before production; and include fine-tuning data provenance.
Reason
Model hubs declare licences in repository metadata and ask users to respect them; licences vary from permissive to non-commercial and use-restricted.
Boundary
Open-weights LLMs, embedding models, pretrained vision models, public datasets.
Symptom or risk
Non-commercial, share-alike, use-restricted or custom licences can forbid the planned use, and finding out after launch forces a rebuild.
Exceptions
Internal research use may be allowed under research licences, but keep it separated from product code.
Trade-offs
Licence review slows experiments; restrict it to candidates heading to production.
Pitfalls
Assuming code licence and weights licence are the same.
Lessons and concepts
Claims and their sources
Hugging Face repositories declare a licence in card metadata, the identifier list includes non-commercial and model-specific community licences, and users are asked to respect a project's licence.