Meta’s Open Source Llama Nearing 350 Million Downloads

TLDR

• Meta said it is seeing close to 350 million downloads of its Llama family of AI models, which is a 10-fold increase from a year ago.
• Meta CEO Mark Zuckerberg the pace is “growing even faster than I expected.”
• Adoption has been broad and rapid, from startups to companies of all sizes, whether they are building in-house or through a cloud provider.

Three years after pivoting to the metaverse along with a name change, Meta is now becoming a juggernaut provider of large language models.

Its flagship Llama family of models is nearing 350 million downloads on Hugging Face, an increase of more than 10x from a year ago, the company disclosed in a blog post.

CEO Mark Zuckerberg said in a Facebook post that adoption of Llama is “growing even faster than I expected.”

By token volume, usage of Llama models has more than doubled from May to July 2024 across the cloud providers where they are available. For some of its largest cloud partners, monthly usage rose by 10x from January to July 2024.

Notably, Meta released its most powerful Llama model in July, Llama 3.1, with a context length of 128K and supporting eight languages. The social media giant said its Llama 3.1 405B is the first open-source AI model on par with frontier-level models from proprietary sources such as OpenAI and Google.

Meta said the adoption of Llama has been “broad and rapid,” from startups to companies of all sizes building by themselves or through a cloud provider. Meta’s cloud partners are AWS, Microsoft Azure, Databricks, Dell, Google Cloud, Groq, Nvidia, IBM watsonx, Scale AI, Snowflake, and others. Future potential partners include Wipro, Cerebras, and Lambda.