<html aria-label="message body"><head><meta http-equiv="content-type" content="text/html; charset=utf-8"></head><body style="overflow-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;">Hi Josh,<div><br></div><div><br></div><div>I would like to share that Meta reportedly refined a new closed source model, Avocado, using Google\u2019s Gemma and Alibaba\u2019s Qwen [1][2]. This led me to consider whether a \u201ccopyleft\u201d model license could help plug this perceived loophole, which is why I initially drafted MG-BY-SA. </div><div><br></div><div>However, it has become clear that such an approach is difficult to make workable in practice, both from a copyright law perspective and from a technical standpoint, as there is no reliable way to verify whether distillation has actually occurred. As a result, I am withdrawing MG-BY-SA for now, as it is likely not OSD-compliant and may not function as I originally intended. </div><div><br></div><div>That said, I remain focused on this issue and will continue to explore whether there is a valid case for MG-BY-SA as a proprietary model license rather than an open source one. My concern is that, in the worst case, large-scale distillation could reduce the willingness of organizations to release open models such as Qwen (under Apache-2.0), pushing them instead toward proprietary service agreements that explicitly prohibit behaviors like distillation. The reverse is also possible, as some models, like DeepSeek and Qwen, have relicensed from custom licenses to OSI-approved licenses.</div><div> </div><div><br></div><div>[1] <a href="https://www.bloomberg.com/news/articles/2025-12-10/inside-meta-s-pivot-from-open-source-to-money-making-ai-model">https://www.bloomberg.com/news/articles/2025-12-10/inside-meta-s-pivot-from-open-source-to-money-making-ai-model</a></div><div>[2] <a href="https://www.itiger.com/news/1143569498">https://www.itiger.com/news/1143569498</a></div><div><br id="lineBreakAtBeginningOfMessage"><div><br></div><div>Best,</div><div>Moming</div><div><br><blockquote type="cite"><div>On Dec 7, 2025, at 03:41, Josh Berkus <josh.berkus@opensource.org> wrote:</div><br class="Apple-interchange-newline"><div><div>All,<br><br>Spinning off a new thread here. I'd like to get feedback from a broader pool.<br><br>Here's the clause in question, present in all three licenses, although it doesn't have the same effect in all three:<br><br>-----------------------------<br>"Derivative Materials" means all improvements, modifications or derivative works to the<br>Licensed Material or any part thereof, which are created or developed by You (either by<br>Yourself or jointly with other third parties), including any derivative model developed by<br>transferring patterns of weights, parameters, activations and/or Output from the Model, such as<br>through distillation methods or synthetic data generation techniques, in order to replicate,<br>approximate, or otherwise achieve functional behavior that is similar to the Model.<br>----------------------------<br><br>This kind of clause is understandible because of the practice of Distillation[1], for example the creation of DeepSeek by training it on ChatGPT [2]. This is a concern for model licensing, since distillation is even being offered as a service these days. This can be regarded by the model owner as a type of copying, although legally it may not be copying.<br><br>This clause raises three questions:<br><br>1. Is this type of clause a violation of the OSD, if the clause does not prohibit distillation, but makes it subject to the conditions of the otherwise-OSS license?<br><br>2. In copyright law, does a clause like this have any possible effect? Or is it cancelled out by the same kinds of precedents that allow LLMs to be trained on copyrighted material while ignoring license conditions?<br><br>3. Does the clause, as written, apply license conditions to distillation effectively?<br><br>4. Does this clause, as written, violate OSD 9 or Freedom 0?<br><br><br>The OpenMDW license[3] does not apply any conditions to model output. However, OpenMDW is also designed as a permissive license, applying minimal conditions to any use of the software. So far, the ModelGo-by--SA license is the first attempt to apply copyleft specifically to models that I know of.<br><br><br>[1]: https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/distillation-turning-smaller-models-into-high-performance-cost-effective-solutio/4355029<br>[2]: https://www.forbes.com/sites/johnwerner/2025/01/30/did-deepseek-copy-off-of-openai-and-what-is-distillation/<br>[3]: https://openmdw.ai/license<br><br>-- <br>-- Josh Berkus<br>OSI Board Member<br></div></div></blockquote></div><br></div></body></html>