Llama’s Current Ecosystem

With over 30 million downloads and more than 7000 related projects on GitHub

Meta released Llama 1 about 7 months ago, and since then Llama 2 and Code Llama have also been released for a few months. On September 27th, local time, Meta published a blog post introducing the current status of the Llama ecosystem. According to official statistics, the Llama model has been downloaded over 30 million times through Hugging Face, with over 10 million downloads in the past 30 days alone.

Cloud providers such as AWS, Google Cloud, and Microsoft Azure have adopted the Llama model on their platforms. Llama 2’s usage in the cloud is expanding. Google Cloud and AWS alone have over 3,500 enterprise projects launched based on the Llama 2 model. Additionally, Meta has announced AWS as the first hosting API partner for Llama 2. Organizations of all scales can access the Llama 2 model on Amazon Bedrock without managing the underlying infrastructure.

Innovative companies, including Anyscale, Replicate, Snowflake, LangSmith, Scale AI, and many others, are among the tens of thousands of startups currently using or evaluating Llama 2. “Innovators like DoorDash are utilizing it for large-scale experiments before rolling out new features supported by LLM.”

Crowdsourced Optimization: As of now, the community has fine-tuned and released over 7,000 derivatives on Hugging Face. On average, they have improved the performance of common benchmark tests by nearly 10% in standard benchmark tests, with performance improvements reaching up to 46% for benchmark datasets like TruthQA.

Hardware Support: Major hardware platforms including AMD, Intel, Nvidia, and Google have improved the performance of Llama 2 through software and hardware optimizations.

Regarding the future of the Llama ecosystem, Meta has stated that they remain committed to providing open approaches for today’s artificial intelligence. They will focus more on multimodality, security responsibilities, and the community moving forward.