NVIDIA adopts Linux Foundation's OpenMDW framework to standardize licensing for its open AI models

We're adopting the Linux Foundation’s OpenMDW framework across our open model families.

This helps make open model licensing simpler and more consistent at scale.

A single legal framework across models, code, documentation, and data helps reduce friction for developers and enterprises building with open source.

Software licenses like Apache 2 were designed for code, not AI models, data, or weights.

The OpenMDW license takes the spirit of successful open software licenses and updates it to protect open technology for AI.

We’ll be using it for many of our Nemotron releases going forward.

this is a huge thing.

i wasn't aware, but @matthew_d_white explained to me that A2.0 and MIT are bad model licenses and that they worked hard with actual lawyers to make such a license but for models

NVIDIA pushing forward means that it has a real shot at adoption now 🤗

Licenses are probably the top very un-sexy things that help the open ecosystem become more long-term stable and economically viable. Excited about this!

@NVIDIAAI onwards !

@NVIDIAAI 🤗

@NVIDIAAI huge leadership for open models. Is hard to be first on these things!

@NVIDIAAI 🙌🙌🙌

Basically this - as much as I love Apache 2.0, it's just not really for models, yanno?

That's why it was important to have the Nemotron Open Model license - which was explicitly designed for that technology.

I'm extremely stoked that there is a new open standard though - I love me an open standard.

@NVIDIAAI this was a big pain point, super nice!

@NVIDIAAI Massive W

@alkimiadev @matthew_d_white huh, you can read the license? there is no discrimination etc.

whether osi wants to approve a non-software license, is on them

@xeophon @matthew_d_white ok so I'm wrong if we're specifically talking about this exact license with exactly nothing else added to it

https://raw.githubusercontent.com/OpenMDW/OpenMDW/refs/heads/main/1.1/LICENSE.OpenMDW-1.1

@xeophon @matthew_d_white Is it actually going to be OSI approved? What I mean is does it allow them to slip in their rhetorical ethical clauses or any other kind of discrimination? I'm not a fan of people doing virtue signaling via licenses.

@rishdotuk @matthew_d_white They’ve done a write up about it! Basically: It covers things like outputs and models aren’t technically software that are covered by A2.0/MIT.

https://www.linuxfoundation.org/blog/the-open-source-legacy-and-ais-licensing-challenge

@NVIDIAAI So great to see OpenMDW both adopted and applied to existing open model families from @nvidiaai .

This is an important step forward in using permissive licensing designed from the ground up for AI models. I hope the industry follows suit.

Great work @nvidiaai team!

@xeophon @rishdotuk @matthew_d_white ty!

@xeophon @alkimiadev @matthew_d_white I didn't find anything objectionable in three reads.

That said, I read it three times because 'by default' made me wonder if they had somehow created affordances for not-permissive variants.

@xeophon @matthew_d_white Why is MIT bad?

@llm_wizard @matthew_d_white I can’t imagine all the work that went into this change, thanks for all the internal champions that fought for this 🥹

Now let’s grow the OpenMDW family 💪