## License incompatibility: MIT License VS BigScience OpenRAIL-M License

#2
by xixi126 - opened

Hi,I'd like to report a license conflict in ldenoue/distilbert-base-re-punctuate. I noticed that this model was quantized from unikei/distilbert-base-re-punctuate, but it's currently published under the mit license. After taking a look at the BigScience OpenRAIL-M License— a license that includes non-permissive terms such as use-based restrictions and attribution requirements. However, this derivative model is currently published under the MIT license, which is very permissive and does not carry over those restrictions.

⚠️ Key conflicts with the BigScience OpenRAIL-M License:

Section III 
4.Redistribution and Derivatives:
  •  Redistribution must include use-based restrictions (Attachment A), which MIT does not require.
  •  Must provide a copy of the BigScience OpenRAIL-M license with any distribution — currently missing.
Attachment A – Use Restrictions:
  •  Prohibits specific uses (e.g., discrimination, surveillance, medical diagnosis, legal decision-making). These restrictions are not enforceable under MIT, which explicitly permits nearly any use.

The MIT license allows:

•  Sublicensing under different terms 
•  Unrestricted commercial use
•  No requirement to pass down upstream ethical or use-based constraints

This creates a clear mismatch: BigScience OpenRAIL-M License imposes enforceable use limitations and distribution conditions that cannot be removed, while MIT explicitly permits those removals.

🔹 Suggestion:

  To comply with the BigScience OpenRAIL-M license terms, it might be helpful to:
  • Include a copy of the BigScience OpenRAIL-M license in the repository or model card
  • Add a notice that the model inherits ethical use restrictions from the upstream model:   
  • Mention that commercial use is restricted, and clarify what uses are not allowed (from Attachment A)
  • Remove the MIT license tag if the full model is not entirely under that license

This would help ensure downstream users are not misled into thinking the model is fully MIT compliant, which it likely is not.

Thanks for your attention!

Looking forward to your response!

I updated the license to use the same. Thanks for noticing this!

Sign up or log in to comment