Start with the OSI definition
Read the Open Source AI Definition and identify its three elements for modifying a machine-learning system.
Start with: The Open Source AI Definition
RedLectern topic
Open and closed weights · open-source definitions · access and usage restrictions.
For: Users · builders · organizations.
After this page you can: Check how a particular model may be used and distributed.
Page reviewed
Part of: Artificial Intelligence › Foundation models
A model’s weights, access method, and permissions are separate questions. Downloadable parameters tell you something about access, while the terms determine what use, modification, and sharing are allowed. The Open Source Initiative (OSI) definition describes four freedoms for AI systems. For machine-learning systems, it names data information, code, and parameters as elements of the preferred form for modifications. It applies the same requirements to a system, a model, weights and parameters, or other structural elements. (Open Source AI Definition)
For a specific model, start with its publisher’s model page and follow the terms linked there. Check whether files are available to download or require an access request, and read the terms for your intended use. Look for conditions on commercial use, modification, attribution, and redistribution. Model-card metadata can display a license or link to custom terms, so follow the link and check the terms for that model. Keep a note of the page and terms you reviewed so you can recheck them later. (Model Cards, Gated models)
Read the Open Source AI Definition and identify its three elements for modifying a machine-learning system.
Start with: The Open Source AI Definition
Start by separating access from permission. Access tells you whether model files are available to you and whether you need to make a request. Permissions come from the terms that accompany the model. The Open Source Initiative’s definition describes the preferred form for modifying machine-learning systems as including data information, code, and parameters. It says the same requirements apply to a system, a model, weights and parameters, or other structural elements. Downloadable weights alone do not establish that a release meets those requirements. (Open Source AI Definition)
For a model you are considering, note whether weights are downloadable, whether access is gated, and which definition a publisher is invoking. Then follow the model’s license or terms link and check whether it covers your intended use.
Read the OSI’s answers about its AI definition.
Learn how a model author can enable access requests for model files.
Open weights
Model parameters available for download; permissions depend on the license.
Closed weights
Model parameters not distributed to users.
Open-source AI
An AI system that meets the OSI definition; downloadable weights alone do not show that it meets the definition.
Model license
Terms governing a model's use, modification, and distribution.
On the Open Source AI Definition, list the three elements it names in the preferred form for modifying a machine-learning system.
The definition names data information, code, and parameters.
A model’s license or other terms may set conditions on using or modifying it. The OSI definition says terms applied to model elements or their combinations may require a modified version to be released under the original terms. (Open Source AI Definition) Do not rely on a short license label alone to answer whether your intended use is allowed.
Open the terms for the specific model and check the activities that matter to your plan: commercial use, modification, attribution, and sharing copies or derivatives. Model-card metadata can display a license and link to a custom license. Follow any such link, and check for separate model-specific terms as well. Record which terms you reviewed so you can recheck them if the model page changes.
Read the OSI’s description of conditions that terms may place on modified versions.
You plan to use a downloadable model in a paid product and share a modified copy. Name three things to verify in its terms.
Check whether commercial use is allowed, what conditions apply to modifications, and whether the original or modified model may be redistributed.
An access path affects what you can obtain and what steps you must complete. Hugging Face’s documentation describes gated models as models whose authors enable access requests. The author can configure automatic or manual approval, and can block access even after a request has been approved. (Gated models) Before requesting access, check what information you will share and what conditions the request asks you to accept.
For a model offered through a service, check the service’s terms alongside any model-specific terms. Using a model through an interface does not itself tell you whether its weights are downloadable or whether copies may be redistributed. Note whether your plan depends on file access, service access, or both, and check the relevant publisher pages.
See how model pages can display license metadata and link to a custom license.
A model repository asks you to request access before downloading. What should you check before submitting the request?
Check what information you will share, whether approval is automatic or manual, which conditions you accept, and whether the author can later block access.
No recent updates are available.
Lawrence Rosen · 2004
A practical foundation for understanding license rights, obligations, and distribution. Its 2004 software focus predates AI model licenses, so check current model terms separately. The author offers the full text online under the Academic Free License 3.0.