Meta is open-sourcing its large language model, Llama 2. The model’s code and weights are being made available free of charge for both research and commercial use.
Llama 2 is the result of the expanded partnership between Meta and Microsoft, with the latter being the preferred partner for the new model. The model is available in the Azure AI model catalog, allowing developers using Microsoft Azure to integrate it into their projects and leverage cloud-native tools for content filtering and safety features. Llama 2 is also optimized to run locally on Windows, creating a less difficult workflow for developers.
According to Meta, Llama 2 has been trained on over 40% more data compared to the previous Llama version and outperforms other language models on reasoning and knowledge tests. Percy Liang, director of Stanford's Center for Research on Foundation Models, commented "LLaMA 2 isn’t GPT-4. but for many use cases, you don’t need GPT-4."
Yann Lecun, chief AI scientist at Meta, said developers interested in leveraging the power of Llama 2 can access the model through Microsoft’s Azure platform, Amazon Web Services (AWS), Hugging Face, and other providers.
Meta says the launch of Llama 2 is accompanied by a number of resources to ensure responsible use. These include red-teaming exercises, a transparency schematic, a responsible use guide, and an acceptable use policy.
Nvidia AI scientist Jim Fan noted "there is a significant gap on coding benchmarks. It's on par or better than PaLM-540B on most benchmarks, but still far behind GPT-4 and PaLM-2-L."
MIT Technology Review noted Meta is not releasing information about the data set that it used to train Llama 2 and the model "has the same problems that plague all large language models: a propensity to produce falsehoods and offensive language."
Developers can refer to the model card for more information or to begin writing their own code. The model card notes among the limitations the model is intended for use in English only.
The move comes after two U.S. senators had previously questioned Meta’s CEO about such releases, writing its "chief AI scientist has stated that open models are key to its commercial success" and asserting the prior open release’s "unrestrained and permissive manner raises important and complicated questions."
Meta has also initiated new programs to gather feedback on the performance and potential improvements of the models. "We believe an open approach is the right one for the development of today’s AI models, especially those in the generative space where the technology is rapidly advancing," Meta said.