Meta and Microsoft have released Llama 2, an artificial intelligence language model for commercial use
Meta and Microsoft have joined forces and introduced Llama 2, a large next-generation language model designed for both commercial and research purposes.
Here's What We Know
Llama 2 is distributed for free. According to Meta, the model was trained on 40% more data than when Llama 1 was created. Contextual length has been doubled.
The release includes model weights and starter code, and the number of parameters ranges from 7 billion to 70 billion. Meta said that Llama 2 outperforms other open source language models on many external metrics, including reasoning, coding, proficiency and knowledge tests.
The development of Llama 2 placed more emphasis on accountability. The developers have red-tested the models and created a transparency scheme that details potential issues.
They also include guidelines for responsible use as well as acceptable use policies to prevent abuses such as criminal activity, misleading representations, and spam.
Microsoft provides the model through the Azure AI catalogue for use in cloud-based tools such as content filtering. The tool can also run directly on Windows PCs and will be available through external providers like Amazon Web Services and Hugging Face.
The first version of the model was open source, but was exclusively for academics and researchers. Llama 2 allows companies to customise the technology for their own purposes, such as creating chatbots or image generators. This allows companies to capitalise on the model and third-party developers to test it for bias, inaccuracy and other flaws.
