Activities / What code does
Trains models
trains_models
Your app uses data to train or fine-tune a machine learning model, rather than only calling a model someone else already trained.
- scope code
- in the registry since 1.0
- 1,610 laws in 256 jurisdictions name it
What triggers it
The lint's definition. It trains, fine-tunes or builds a model on data it collects. Calling somebody else's model is not this.
This covers building a recommendation model from user activity, fine-tuning a language model on your own documents, and training an image classifier on photos users upload.
What does not
Calling a third-party model's inference API without training or fine-tuning anything does not count.
Often confused with Generates content with AI and Crawls the web.
The law it brings into scope
The kinds of law that name this activity in the LexLint law library, with the number of laws and jurisdictions in each, as of . The examples are chosen by rule, laws in force first and the widest jurisdiction first; each section's page lists every law.
Counterparties that usually come with it
- a cloud platform that runs the training
- a dataset host or data broker that supplies the training data
A component that cannot run as documented without one of these names
it in its ACTIVITIES.md under talks_to, with
required: true. That party's jurisdiction then joins the
run of every product that uses the component, and the finding says
why it is there. A default the deployer can change is asked about,
never added. Talking to a party never makes a product that party.
In a declaration
A library or framework that makes this possible enables
it; an application whose shipped code does it performs it.
For a library
For an application
History
In the registry since version 1.0, .
In the glossary: Trains models.
The activity registry is LexLint's open list; this page's text is licensed under Creative Commons Attribution 4.0.