Meta has never released an Open Source model, so I don't think they're interested in that.
Actual Open Source base models (all Apache 2.0 licensed) are Falcon 7B and 40B (but not 180B); Mistral 7B; MPT 7B and 30B (but not the fine-tuned versions); and OpenLlama 3B, 7B, and 13B.
You can tell Meta are well aware of this by the weasely way they use "open" throughout their marketing copy. They keep talking about "an open approach", the document has the word "open" 20 times in it, and "open source" once where they say
Aligned with our open approach we look forward to partnering with the newly announced AI Alliance, AMD, AWS, Google Cloud, Hugging Face, IBM, Intel, Lightning AI, Microsoft, MLCommons, NVIDIA, Scale AI, and many others to improve and make those tools available to the open source community.
which is obviously not the same as actually open sourcing anything. It's frustrating how they are deliberately trying to muddy the waters.
The Llama 2 model license requires agreeing to an acceptable use policy, and prohibits use of the model to train competing models. It also prohibits any use by people who provide products or services to more than 700M monthly active users without explicit permission from Meta, which they are under no obligation to grant.
These restrictions violate terms 5 (no discrimination against persons or groups) and 6 (no discrimination against fields of endeavor) of the Open Source Definition.
Actual Open Source base models (all Apache 2.0 licensed) are Falcon 7B and 40B (but not 180B); Mistral 7B; MPT 7B and 30B (but not the fine-tuned versions); and OpenLlama 3B, 7B, and 13B.
https://huggingface.co/tiiuae
https://huggingface.co/mistralai
https://huggingface.co/mosaicml
https://huggingface.co/openlm-research