Meta Releases Llama 2 AI Model for Free Commercial Use
Meta open-sources its Llama 2 large language model, making it available for both research and commercial applications through a partnership with Microsoft. The updated model trains on significantly more data than its predecessor while including new resources for responsible AI use.
Meta releases its Llama 2 large language model as an open-source tool, providing the code and weights free of charge for both research and commercial use. This launch stems from an expanded partnership with Microsoft, which serves as the preferred partner for the new model and hosts it in the Azure AI model catalog. Developers easily integrate Llama 2 into their projects using Microsoft's cloud-native tools for content filtering and safety features.
The updated model trains on over 40% more data than the previous Llama version and outperforms other language models on various reasoning and knowledge tests. Experts note that while Llama 2 does not match the capabilities of GPT-4, it provides more than enough power for many common use cases. The model is also optimized to run locally on Windows, which simplifies the workflow for developers building applications outside of cloud environments.
Meta accompanies the launch with several resources designed to ensure responsible use, including red-teaming exercises, a transparency schematic, and an acceptable use policy. However, the model still shares common large language model limitations, such as a tendency to produce falsehoods and offensive language, and currently supports English only. Developers access Llama 2 through Microsoft Azure, Amazon Web Services, Hugging Face, and other major providers.