Meta has released a large language model called Llama 2, which the company claims is open source. The license for Llama 2 however contains restrictions that prevent it from meeting the definition of open source set by the Open Source Initiative (OSI).
Specifically, the license forbids the use of Llama 2 to train other language models, and it requires a special license from Meta if the model is used in an app or service with more than 700 million monthly users. These restrictions mean that Llama 2 is not truly open source, as it does not allow for unlimited distribution and use of the software, and it places restrictions on the field of use.
The OSI’s definition of open source requires that the software must be distributed in source code form, so that users can modify and redistribute it. The license must allow for unlimited distribution and use of the software, with or without modification, and that the license must not place any restrictions on the field of use of the software.
Llama 2’s license does not meet these criteria, as it forbids the use of the model to train other language models and requires a special license for use in large-scale applications. As a result, Llama 2 is not truly open source.
The OSI has criticized Meta’s licensing terms, stating that they “do not comply with the Open Source Definition.” The OSI has also called on Meta to release Llama 2 under a more permissive license that would allow for its unrestricted use.
The sources for this piece include an article in TheRegister.