Meta Says Its LLaMA AI for Researchers Does More with Less

Meta Platforms has unveiled a new generative artificial intelligence language system called LLaMA, which doesn’t chat, but is designed as a research tool the company hopes will help “democratizing access in this important, fast-changing field.” The LLaMA (Large Language Model Meta AI) ranges in size from 7B to 65B parameters. Touted as a “smaller, more performant model,” LLaMA enables those members of the research community that do not “have access to large amounts of infrastructure to study these models,” Meta explains. Training smaller foundation models requires less computing power and resources for testing and validation.

Foundation models train on a large set of unlabeled data, which Meta says “makes them ideal for fine-tuning for a variety of tasks.” In addition to 7B and 65B, LLaMA will be available in 13B and 33B parameters. Meta says it will be sharing a model card with details on how LLaMA is “in keeping with our approach to Responsible AI practices.”

LLaMA will be available to researchers, academia, civil society and entities affiliated with government under non-commercial license, Meta said in a blog post that emphasizes the system’s use to researchers who identify and address biaes and inaccuracies of popular AI language models like ChatGPT.

“LLaMA-13B outperforms GPT-3 (175B) on most benchmarks,” Meta said in an abstract to a scholarly paper that claims “LLaMA-65B is competitive with the best models including DeepMind’s Chinchilla 70B and Google’s PaLM-540B.”

“Once trained, LLaMA-13B can also run on a single data center-grade Nvidia Tesla V100 GPU,” writes The Verge, noting that will be “welcome news for smaller institutions wanting to run tests on these system but doesn’t mean much for lone researchers for whom such equipment is out of reach.”

Earlier this month, Meta introduced Toolformer, an AI platform that has the ability to teach itself to use external apps and APIs. Also in February, Microsoft announced an ChatGPT-powered version of its Bing search engine, while Google unveiled Bard, it’s own version of chat-based AI-driven search.

Last week, China’s Baidu announced plans to integrate its Ernie chatbot into all services, including search, cloud, smart car OS and smart speakers, according to The Washington Post.

Shares of Baidu “surged by more than 13 percent in Hong Kong” on the the Ernie news, says CNBC, exemplifying the global appetite for AI chat.

No Comments Yet

You can be the first to comment!

Leave a comment

You must be logged in to post a comment.