<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,<div><br></div><div><br></div><div>Thanks for Pam\u2019s insightful comments.</div><div><br></div><div>Actually, I do not insist on including models created through distillation in MG0 and MG-BY, because these licenses are intended to be permissive. Although I personally believe there may be a licensing gap that could help balance behaviors such as distillation, I agree that this should not be addressed within an Open Source license.</div><div><br></div><div>My concerns about potential over-broad interpretations of the current definitions also align with Pam\u2019s view that output-related materials should not be considered derivative in the copyright context. A more suitable approach might be to introduce a new definition such as \u201cDistillation Materials\u201d and use that term only in non\u2013Open Source licenses like MG-BY-NC. Many well-known model licenses (e.g.,OpenRAIL, AI2 ImpACT, Gemma) use similar definitions, but I agree that it is not reasonable to follow these definitions within an Open Source model license, and it is better to draw a clear line between them.</div><div><br></div><div>The main gap I intend MG0 and MG-BY to fill is simply to give developers a license option when they want to contribute their models to benefit the community without asking for anything in return. As my yoga trainer says, \u201cIt is more blessed to give than to receive.\u201d This is particularly true in the current ML community. For example, you can find that almost all the most popular models on Hugging Face (https://huggingface.co/models) have only a single (initial) commit, downstream users create derivatives (view in the \u201cModel Tree\u201d) to fit their own scenarios without contributing back (and there is little reason to do so, since their derivatives are already customized and not as general-purpose as the base models). This is quite different from the contribution patterns seen in software projects.</div><div><br></div><div><br></div><div>In short, I will remove the following definition:</div><div><br></div><blockquote style="margin: 0 0 0 40px; border: none; padding: 0px;"><div>\u201cincluding any derivative model developed by transferring patterns of weights, parameters, activations and/or Output from the Model, such as through distillation methods or synthetic data generation techniques, in order to replicate, approximate, or otherwise achieve functional behavior that is similar to the Model.\u201d </div></blockquote><div><br></div><div>I will also update MG0 and MG-BY according to the latest comments from McCoy. The updates will not take long, and I will be back soon.</div><div>Regarding MG-BY-SA, I intend to withdraw it for the following reasons: </div><div><br></div><div>1) I don\u2019t see a critical gap that this license currently fills (especially after removing the distillation-related definitions). As I mentioned earlier, models are unlikely to be contributed back to the main branch, and there is little motivation for either publishers or users to do so.</div><div><br></div><div>2) By removing distillation from the definition of derivatives, I am concerned that this license will not function as some might expect for a ShareAlike-style model license. </div><div><br></div><div>3) I believe that copyleft model license topics require more thoughtful discussion and time. </div><div><br></div><div><br></div><div>Also, thanks to Josh for creating a new discussion thread about distillation and copyleft model licenses. I think this is one of the most valuable topics in ML licensing, and I am looking forward to continuing the discussion.</div><div><br></div><div><br></div><div><br></div><div>Best,</div><div>Moming</div><div><br></div><div><br id="lineBreakAtBeginningOfMessage"><div><br></div></div></body></html>