Meta Launches Llama 2 for Commercial Apps
- By Paul Mah
- July 19, 2023
Meta this week announced the next version of its open-source AI language model. And unlike its predecessor, the new Llama 2 family can be integrated into commercial products.
Bigger and faster
According to the Llama 2 website, the new models range in size from 7 to 70 billion parameters and are pretrained on 2 trillion tokens with a context length of 4,096. A longer context length gives the model more information and a richer context for generating a response.
LLaMa 2 was trained on 40% more data compared to Llama 1, and outperforms open source chat models “on most benchmarks” it tested, says Meta.
“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,” wrote Meta on a blog.
“And we believe it’s safer. Opening access to today’s AI models means a generation of developers and researchers can stress test them, identifying and solving problems fast, as a community.”
A report on Ars Technica cited a senior AI scientist at Nvidia who says Llama 2 is close to GPT-3.5 on reasoning tasks, albeit not for coding, and is “on par or better” than PaLM-540B on most benchmarks.
On the same day, Qualcomm announced it is working with Meta to bring Llama 2-based AI implementations to laptops, smartphones, vehicles, and IoT devices starting from 2024. This will offer a new generation of on-device AI that doesn’t rely on cloud services.
Though Llama 2 is open source, it is worth noting that Meta did not disclose the source of the training data used to create the models, as noted by Abeba Birhane, a senior fellow in Trustworthy AI at Mozilla.
Llama 2 is currently available in the Azure AI model catalog on Microsoft Azure, Amazon Web Services (AWS), Hugging Face, and other providers, and is also optimized to run locally on Windows.
Additional details about Llama 2 are available on the research paper released by Meta here.
Paul Mah is the editor of DSAITrends. A former system administrator, programmer, and IT lecturer, he enjoys writing both code and prose. You can reach him at [email protected].
Image credit: Meta
Paul Mah
Paul Mah is the editor of DSAITrends, where he report on the latest developments in data science and AI. A former system administrator, programmer, and IT lecturer, he enjoys writing both code and prose.