Hugging Face OpenLLM Leaderboard

IBM Unveils Open-Source Granite 3.0 AI Models with 2B and 8B Parameters at TechXchange

Oct 21, 2024, 12:35 PM

IBM has unveiled its new open-source Granite 3.0 family of AI models at the company's annual TechXchange event, designed specifically for enterprise applications. Released under the Apache 2.0 license, the Granite 3.0 lineup includes both dense and mixture-of-expert architectures with 2-billion and 8-billion parameter models. These models are trained on over 12 trillion tokens across multiple languages and programming languages, aiming to deliver high performance and efficiency in enterprise AI tasks. IBM claims that the Granite 3.0 models outperform similarly sized models, such as Llama-3.1 8B, on benchmarks like the Hugging Face OpenLLM Leaderboard (v2). The models are optimized for CPU-based deployments and edge computing, highlighting IBM's commitment to providing accessible and cost-effective AI solutions for businesses. The Granite 3.0 models are part of IBM's Watsonx platform and are linked to IBM's Consulting Advantage, offering enhanced AI capabilities to enterprise clients.