This is not an open source project

[richard-delorenzi]

CC BY-NC-SA 4.0 (Creative Commons Attribution-NonCommercial-ShareAlike 4.0) is not an open-source or open-content license because of the "NonCommercial" (NC) restriction. While it allows for sharing and adapting, it fails the definition of "open" by restricting commercial usage, which is a requirement for true open-source, open-definition, or free-culture works.

You can read the Open Source definition. Or you can look at approved licences.

Problematic Aspects of CC BY-NC-SA 4.0:

• NonCommercial (NC): Limits usage to non-commercial activities.
• Not for Software: Creative Commons advises against using this for software.

What are true CC open licenses?

Only CC BY and CC BY-SA meet the standards of the "Open Definition" and open access standards, as they allow for commercial

However, none of the CC licences are recommended for software.

[nocnestudio]

Open and Try Only... ? What next... ?

[milasudril]

See also #68

[nikopueringer]

I understand where you guys coming from regarding "true" open source.

Here are my goals: I want the project to be free to use for any (legal) purpose. The code is may be modified and redistributed as long as it remains free. The model and inference code may not be packaged into commercial software without Corridor Digital's explicit permission. Any contributions to this repo become property of Corridor, which is currently stated in the license, but i want to make sure people are aware, understand the intent, and are okay with it. The project is intended to be communally developed and freely useable and accessible, and we are specifically controlling people's ability to commercialize the project in order to protect this mission. We are also making a point to be up front and clear in our mission and goals, so that people are fully aware should they choose to get involved.

I have seen a few suggestions that I look into the GPL family of licenses. I'll be doing that shortly here. I am open to other ideas though. Any other suggestions I should look in to?

[magikmw]

GPL would work great for this - in particular, GLP-3.0 might fit your vision here, but the less restrictive GPL-2.0 is also a good choice (Linux Kernel uses it exclusively, and Torvalds talked about why several times).

I'd be happy to help if you have any particular questions about licensing.

[anders-d]

The main problem in my view is that it is Meta's Hiera model that does the "heavy lifting" in this project, and the checkpoint used is fine-tuned from the "facebook/hiera_base_224.mae_in1k_ft_in1k" base weights. Those base weights are licensed by Meta/facebookresearch with CC BY-NC 4.0, and Corridor's fine-tuned weights inherits that license, and can therefore not be used for commercial purposes.

see:
#68
#113
#2
#152

[magikmw]

The main problem in my view is that it is Meta's Hiera model that does the "heavy lifting" in this project, and the checkpoint used is fine-tuned from the "facebook/hiera_base_224.mae_in1k_ft_in1k" base weights. Those base weights are licensed by Meta/facebookresearch with CC BY-NC 4.0, and Corridor's fine-tuned weights inherits that license, and can therefore not be used for commercial purposes.

Good point. However, if training data and weights get released - those can be licensed differently. As long as they'd be used with another model as a basis, that wouldn't interfere with Corridor's right to license however they like.

[richard-delorenzi]

I think the first step is to stop claiming that it is open-source. Then look at the Gnu licence that protects code run on servers. I don't remember which one, as am no-longer doing the legal side.

[milasudril]

@richard-delorenzi While the GPL focuses on traditional distribution (e.g., shipping binaries or source code), the AGPL ensures that even if users interact with modified software remotely, the provider must share the source code of those changes.

GPL v3 protects against so-called tivoization. That is, running the software on locked-down systems.

[tuomosipola]

I could see GPL working well with the stated goals. It is a bit like CC-BY-SA but for code. If people creating their own services with the code seems like a problem, AGPL might be an alternative then (requires code sharing for online services). Finally, licensing the code and data with different licenses is a possibility, if Hiera license is a problem. Users could download the data separately.

[nikopueringer]

Thanks for all the feedback! Will be continuing to work through this. Also might be training next model from scratch, so no Hiera licensing to worry about. In the meantime you guys have helped me better understand which licenses I may need to turn to.

[henryruhs]

Good point. However, if training data and weights get released - those can be licensed differently. As long as they'd be used with another model as a basis, that wouldn't interfere with Corridor's right to license however they like.

That is nonsense, you inherit the license from the pretrained weights used via training.

[storreira]

I’ve been following this discussion, and it touches on a critical point for the future of AI-driven legal and software engineering: the gap between 'Source Available' and 'Open Source'.

Nikopueringer, your goal—to ensure the project remains free and prevent third-party commercialization without permission—is valid, but as others have pointed out, CC BY-NC-SA 4.0 effectively removes the project from the Open Source ecosystem. This isn't just a semantic issue; it creates compliance friction:

Legal Uncertainty in Commercial Contexts: In jurisdictions like Argentina (where I practice), 'commercial activity' is interpreted broadly. Even a non-profit using your code to optimize internal legal workflows could be seen as a violation, scaring away professional contributors.

The 'Hiera' Dependency: If the weights are fine-tuned from Meta’s CC BY-NC 4.0 base, the project is legally a 'Derivative Work'. You cannot license the output more permissively than the input allows. Training from scratch is indeed your only path to licensing sovereignty.

The Solution for your Intent: If you want to protect the 'mission' while being truly open, the AGPLv3 (Affero GPL) is likely your best bet. It closes the 'SaaS loophole': if anyone runs your code on a server to provide a service, they must share their modifications. It doesn't ban commercial use, but it forces 'Copyleft' transparency, which usually deters proprietary exploitation better than an NC clause.

Strictly speaking, you are currently in Source Available territory. Transitioning to AGPLv3 (once you have clean weights) would align you with the Open Source Definition while keeping the community-driven 'protection' you're looking for.