state-of-open-source-ai/models/

[utterances-bot]

Models — State of Open Source AI Book

https://book.premai.io/state-of-open-source-ai/models/

[flaxsearch]

Llama 2 is not open source, the Meta licence isn't OSI-approved. Sadly Meta keep saying it is open source and people keep believing them.

[biswaroop1547]

@flaxsearch True, thanks for pointing out! hence we also mentioned -

All model variants under LLaMA-2 are released under LLaMA-2 License, permitting commercial usage unless it’s facing 700 million monthly active users then the entity must obtain a license from Meta.

[casperdcl]

See also Meaning of "Open" - I agree it's deliberately confusing. Open source weights doesn't have to mean open source training data or permissive/OSI-approved licence terms.

[flaxsearch]

Perhaps you should retitle the section 'Open Source Models' as 'Open Models' and then link to the section on Meaning of Open just below the title? I agree it's confusing, I wrote https://opensourceconnections.com/blog/2023/07/19/is-llama-2-open-source-no-and-perhaps-we-need-a-new-definition-of-open/ in an attempt to help clarify the situation

[casperdcl]

Good idea; added a link to Meaning of "Open" in #97

Also note that OSI's "open source definition" (OSD) is mentioned in the link above, but I completely disagree with it. OSD states that "open source" in their opinion should also imply "open licence", and it focuses almost exclusively on licences rather than source code. This is wrong. Source code and licences are two independent, well-defined concepts and do not at all need to imply each other. I believe OSD is the biggest contributor to confusion, and I would strongly argue that OSD should be renamed "open licence definition".

For me a more interesting point is "can you really call a model open source if only the weights but not the training data are available?"