[License-review] ModelGo Attribution-OpenSource License, Version 2.0

Mon Mar 3 02:14:19 UTC 2025

Dear Moming,

First off, thank you for submitting the licenses. I've attached a .odt 
version of the BY-OS license with extensive comments. This is more about 
using your license as a teaching opportunity than directed at your 
licenses specifically. Since you plan to revise and resubmit the 
licenses, I took advantage of an opportunity to provide some suggestions 
on improvements. They are intentionally very detailed, some admittedly 
picayune, because I am using them to educate more broadly.

Open source licenses are quite different from ordinary commercial 
licenses. Open source licenses are meant to give away as much as 
possible, with only a few conditions that the licensor requires as their 
benefit from granting the license (attribution, copyleft), and some 
protection from claims.

On the other hand, commercials licenses are designed to grant as few 
rights as possible and shift as much of the risk and burden to the 
licensee as the licensor can get away with. For that reason 
stylistically they are quite different from the open source license, 
where there is not generally a shift of risk or burden but just 
protection for the licensor. Think of the open source license as a gift 
- when you give a gift you protect yourself ("I'll give you the car but 
you have to register it in your own name") but when you start to impose 
too many burdens on the gift recipient ("you have to pay me $50 a week 
or I'll take the car back") it's no longer a gift. Open source as a 
practice is successful only because licensees know that they can used 
the software (or model, in this case) freely without worrying about what 
limits there might be on their use because there are only very few, 
well-known limits. Imposing too many burdens and obligations on the 
licensee means that benefits of the open source ecosystem won't be realized.

Another difference is that commercial licenses are low risks as 
documents. They are between only a few parties; even a click-through 
license pales in the number of licensees compared to an open source 
license. We now know that open source licenses have lifespans of decades 
but commercial licenses generally don't. The commercial licensor will 
generally be able to modify and improve their license over time as they 
discover weaknesses or ambiguities, but that doesn't happen in open 
source licenses because they are perpetual. In open source one might be 
able to change a license for a later version of the software (depending 
on whether the open source licensor own all the rights), but anyone who 
has either updated a license version or changed licenses will tell you 
what a difficult, multiyear process it is. This means that the open 
source license can't be sloppy; it has to be as perfect as possible.

The ModelGo licenses were written in the style of commercial licenses 
and so have terms that perhaps shouldn't be included an open source 
license. I've also taken the liberty to point out many areas with 
imprecise drafting that creates interpretive problems. To the extent 
poor drafting creates ambiguity about whether all necessary rights are 
granted, it will mean that, in my view, the license shouldn't be approved.

In that category, unintended ambiguities that I think are fatal to the 
license are:

Definition of Licensed Materials -- you say "'Licensed Materials' means, 
collectively, the Model and/or Complementary Materials ...." Are you 
saying that what is licensed under this license must include both the 
Model AND the Complementary Materials (as indicated by the word 
"collectively") or can the license be applied to only a Model (as 
indicated by the words "and/or")? I don't know if this license is meant 
for everything for an AI system except the data or it can be used for a 
model alone without all the supporting materials. This ambiguity about 
what is being licensed creates a lot of interpretive problemes. Say for 
example that I receive both the Model and Complementary Materials but I 
want to Distribute only the Model - is it ok that I don't give my 
distributees all the materials I received? Or must I Distribute the full 
package? If it's ok that I distribute the Model only, do I still need to 
provide the Source Code for the Complementary Materials? If I create 
both a new model and have new Complementary Materials, but I plan to 
distribute only my new model as the Derivative Materials, do I have to 
nevertheless have to provide my own Complementary Materials?

2.1(a) -- you grant a license to "worldwide right and license (including 
the relevant copyrights and patent rights) ..." However, you have 
defined "Intellectual Property Rights" more broadly than patent and 
copyright, so by negative implication you appear not to have granted all 
rights that the user might need, i.e., you have not granted Intellectual 
Property Rights that aren't copyright or patent.

2.1(a)(i)-(ii) -- as worded, it sounds like you are NOT granting rights 
to use, reproduce and Distribute Derivative Materials, so you are not 
granting all rights needed for someone to exercise all rights required 
for an open source license. I think there is an argument around it 
(i.e., romanette (i) grants the necessary rights for any Licensed 
Materials that are embodied in the Derivative Materials, and the 
Licensor can't grant any rights for the portion of the Derivative 
Materials the Licensor did not create because the Licensor doesn't own 
them, its the person who created the Derivative Materials who has to 
grant the license for their share of the Derivative Materials), but it 
could be worded more clearly.

2.5(a) -- you have reserved rights that may be needed for someone to 
exercise the full grant required for an open source license, in 
particular because you expressly aren't granting rights in database. One 
theory for the protection of models is database rights, so you might 
have unintentionally withheld the right to reproduce the Model.

I'm not suggesting that all of my comments are right or you have to 
adopt them. My suggestions may also be inconsistent; making sure 
licenses are internally consistent is quite time consuming. License 
drafting is also iterative; you have to see changes before you spot the 
problems the new changes created or exposed. So don't take these 
suggestions as absolute truth or necessary, but food for thought about 
how to draft a higher quality open source license.

Pam

Pamela S. Chestek (in my personal capacity)
Chestek Legal
4641 Post St.
Unit 4316
El Dorado Hills, CA 95762
+1 919-800-8033
pamela at chesteklegal
www.chesteklegal.com

On 2/12/2025 5:26 AM, Moming Duan wrote:
> Dear OSI Community,
>
>
> I am Moming Duan, a researcher at the National University of Singapore 
> and the submitter and license steward of 
> *ModelGo Attribution-OpenSource License 2.0*, which I am submitting 
> for OSI review through this email. The license TEXT file is attached, 
> and below is a brief overview of this license.
>
> *License Name*:ModelGo Attribution-OpenSource License
> *Version*: 2.0
> *Short Identifier: *MG-BY-OS-2.0
> *Copyleft:*No
> *Legacy or New*: New License
> *Drafted By Lawyer*: Yes, Rajah & Tann Singapore LLP
> *Approved or Used by Projects*: No
>
> *License URL*:https://ids.nus.edu.sg/modelgo-mg-by-os.html
> *Introduction and Video*:https://www.modelgo.li/
>
> *Overview*:
>
> ModelGo Attribution-OpenSource License Version 2.0 (MG-BY-OS-2.0) is a 
> new license designed for publishing models (typically neural networks 
> like Llama2, DeepSeek). It is one of the variants in the ModelGo 
> License family. MG-BY-OS-2.0 is the a copyleft license in the ModelGo 
> family, requiring that the original license and attribution be 
> provided when distributing the original Licensed Materials or 
> Derivative Materials (Licensed Materials and Derivative Materials 
> aredefined in Clause 1.1). A statement of modification is required, if 
> applicable. Derivative Materials should be licensed under the same 
> terms as MG-BY-OS-2.0, and redistribution of original works or 
> derivatives should include the source code. This license is intended 
> to be an open-source model license that provides as much openness as 
> possible within the scope of the model itself (in contrast to Llama2 
> license and OpenRAIL licenses). While it is not a determining factor 
> for an open-source AI system, it can be considered one of its 
> requirements.
> (Green content represents the differences from MG-BY-2.0 license)
>
> *Complies with OSD:*
> *
> *
> OSD 3 Derived Works — MG-BY-OS-2.0 Clause 2.1 (a) grants copyright and 
> patent rights to create derivatives.
> OSD 5 and OSD 6 — No discrimination clause is included in MG-BY-OS-2.0.
> OSD 9 License Must Not Restrict Other Software — No such restriction 
> is included in MG-BY-OS-2.0.
>
> *The Gap to Fill:*
> Model sharing is very common on the web, with over 1.4 million models 
> currently listed on Hugging Face (https://huggingface.co/models). 
> However, most of these models are not properly licensed. When 
> publishing their models, developers typically choose from three main 
> options (as seen in the model license tags on the Hugging Face website):
>
>   * OSS licenses, e.g., Apache-2.0, MIT
>   * Open responsible AI licenses (OpenRAILs),
>     e.g., CreativeML-OpenRAIL-M, OpenRAIL++
>   * Proprietary Licenses, e.g., Llama2, Llama3
>
>
> However, not all licenses are well-suited for model publishing.
>
> *Why not use OSS licenses? *
> Traditional OSS licenses lack clear definitions regarding machine 
> learning concepts, such as Models, Output, and Derivatives created 
> through knowledge transfer. This ambiguity can result in certain ML 
> activities (e.g., Distillation, Mix-of-Expert) being beyond the 
> control of the model owner.
>
> *Why not use OpenRAILs? *
> Recently, Responsible AI Licenses (https://www.licenses.ai/) have been 
> widely advocated to govern AI technologies, aiming to restrict 
> unlawful and unethical uses of models. While I acknowledge the growing 
> need for such governance, these copyleft-style restrictions do not 
> comply with the OSD and may cause incompatibility with licenses like 
> GPL-3.0. Another concern is that these behavioral restrictions may 
> proliferate within the AI model ecosystem, increasing the risk of 
> license breaches.
>
> *Why not use Llama2 or Llama3 Licenses?*
> These licenses are proprietary licenses that are not reusable. 
> Furthermore, they include exclusive terms such as "You will not use 
> the Llama Materials or any output or results of the Llama Materials to 
> improve any other large language model" and copyleft-style behavioral 
> restrictions.
>
> In fact, the dilemma in current model publishing is the lack of a 
> general-purpose license for model developers. Additionally, since no 
> single license meets diverse model publishing needs, some developers 
> resort to using CC licenses with different elements. However, CC 
> licenses are ill-suited for this purpose as they do not grant patent 
> rights. This motivated the drafting of ModelGo License family, which 
> provides different licensing elements similar to CC but specifically 
> designed for model publishing.
>
> *Comparison with Existing OSI-Approved Licenses:*
> Since I could not find an OSI-approved model license, I can only 
> compare MG-BY-OS-2.0 with one similar OSS license — Apache-2.0
>
> # MG-BY-OS-2.0 defines licensed materials and derivative works 
> differently from Apache-2.0, tailoring them to models.
> # MG-BY-OS-2.0 Clause 2.4 includes provisions regarding model output.
> # MG-BY-OS-2.0 Clause 2.2(a) clarifies the ownership of Derivative 
> Materials.
> # MG-BY-OS-2.0 Clause 7 specifies the governing law.
> # MG-BY-OS-2.0 Annex A includes a Model Sheet to help users choose and 
> understand the license content.
> # MG-BY-OS-2.0 can govern the remote access (e.g., chatbot) scenario.
> # MG-BY-OS-2.0 is a copyleft license and requires the source code to be 
> provided during redistribution.
>
> If further comparisons or supporting evidence are needed to strengthen 
> my claims, please let me know. I am more than willing to engage in 
> further discussions with the OSI community about this license and 
> contribute to promoting standardized model publishing. 🤗
>
>
> Best,
> Moming
>
>
>
> _______________________________________________
> The opinions expressed in this email are those of the sender and not necessarily those of the Open Source Initiative. Communication from the Open Source Initiative will be sent from an opensource.org email address.
>
> License-review mailing list
> License-review at lists.opensource.org
> http://lists.opensource.org/mailman/listinfo/license-review_lists.opensource.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opensource.org/pipermail/license-review_lists.opensource.org/attachments/20250302/f582183c/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: MG-BY-OS-2.0 redlined.odt
Type: application/vnd.oasis.opendocument.text
Size: 72851 bytes
Desc: not available
URL: <http://lists.opensource.org/pipermail/license-review_lists.opensource.org/attachments/20250302/f582183c/attachment-0001.odt>