your123 - stock.adobe.com
Meta's second generation of its large language model, Llama 2, gives enterprises something they can test and work with.
The social media and technology vendor has introduced Llama 2 as an open source model free for research and commercial use. The July 18 release of Llama 2 also included model weights and starting code for both the pre-trained and fine-tuned versions of the model.
Meta also revealed that Microsoft is its preferred partner for Llama 2 and that the LLM is available in the Azure AI model catalog. It is also optimized to run locally on Windows. The LLM is now available through other providers, including AWS and Hugging Face.
Open source models
The new generation of Llama follows the expected trajectory of the open source model, said Mark Beccue, research director at The Futurum Group.
"We're getting used to these things being updated quite a bit," Beccue said. "I don't see anything that tells me it's any better or worse than anything else."
However, Meta's choice to make Llama 2 open source for researchers and enterprises makes the LLM market more competitive, Beccue added.
"Open source is really important to the whole ecosystem because that keeps everybody else on their toes," he said.
Llama 2 is especially important for enterprises because they can build something with it compared to the first generation, which was only available to researchers, Forrester Research analyst Rowan Curran said.
"Having these models be out there and open source means that there's a lot more people who are working on them, who are exploring with them and who can understand what their best applications and limitations are," he said.
Making Llama open source in the commercial arena is in contrast with what competitors OpenAI, Alphabet and others are doing.
Mark BeccueResearch director, Futurum Research
"While today the closed sourced models -- such as the ones from OpenAI, Microsoft, Google -- have performance advantage, Llama 2 has the potential to narrow that gap between closed and open source models," Gartner analyst Arun Chandrasekaran said.
It's possible that despite its partnership with Meta's rival OpenAI, Microsoft is taking advantage of enterprises' interest in open source models, he added.
"Microsoft perhaps saw an opportunity to offer infrastructure to train and run these models as well as provide operational tools for deploying them in a robust and safe manner," Chandrasekaran said.
Meta also released Llama 2-chat, a fine-tuned version of the model trained for two-way conversations.
Esther Ajao is a news writer covering artificial intelligence software and systems.