Can pre-trained models be used in commercial applications?

[Franck-Dernoncourt]

https://twitter.com/ylecun/status/1629189925089296386 (mirror 1, mirror 2, mirror 3) says yes (with the GPL v3 license):

Meta is committed to open research and releases all the models the research community under a GPL v3 license.

https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md says no:

License Non-commercial bespoke license.

So I'm confused.

[nightvision04]

Very important question right here.

[dionator]

Since the tweet got deleted right after I opened this issue, here's a copy:

The deletion makes me think it's not GPL v3 license though.

The LinkedIn Post claiming GPL v3 is still on though: https://www.linkedin.com/feed/update/urn:li:activity:7034956639526952960/

[Franck-Dernoncourt]

@dionator my bad, typo. Fixed.

[MesiA]

I think this issue should be closed: https://twitter.com/ylecun/status/1629214720405143552

[dizys]

I think this issue should be closed: https://twitter.com/ylecun/status/1629214720405143552

My guess is that the model architecture code is open sourced under GPL3, but the pretrained weights ain't. Technically Yann isn't wrong, just that he might not be referring to the trained weights which are gated and not in this repo.

[andrewssobral]

Hello everyone,

Has anyone successfully obtained the pre-trained weights? I submitted the form a few days ago, but haven't received a response yet.

[nightvision04]

Not yet

[subramen]

Llama2 is permissible for commercial use with an important caveat: https://github.com/facebookresearch/llama/blob/main/LICENSE#L65

[Franck-Dernoncourt]

Thanks! Another important caveat of Llama 2, from the license, which also impacts non-commercial work/research:

v. You will not use the Llama Materials or any output or results of the Llama Materials to improve any other large language model (excluding Llama 2 or derivative works thereof).