AI21, a company based in Tel Aviv, Israel, announced a new collaboration with Microsoft to integrate their Jamba model into Microsoft Azure AI. This collaboration brings AI21’s latest Jamba-Instruct model to Azure AI as part of its Models as a Service (MaaS) offering. The Jamba-Instruct model is the first production-level LLM based on the Mamba-Transformer hybrid architecture, which AI21 introduced to the market. With a context window of 256K, Jamba-Instruct sets a new standard for enterprise-grade AI solutions.
Traditional LLMs typically use the Transformer architecture, which has challenges such as high memory usage and slow inference with growing context, limiting practical production use. Through this collaboration, Azure customers will have access to Jamba-Instruct without the need to manage the underlying infrastructure, making Azure the first hyperscaler cloud to offer this model to customers. This integration highlights Microsoft’s commitment to providing advanced AI tools and technologies for businesses to drive innovation and digital transformation.
Jamba-Instruct is designed to deliver value, quality, and performance for enterprise applications. With a large context window capable of processing the equivalent of a 400-page novel, Jamba-Instruct excels at tasks like question answering, document summarization, and creating chatbots for sustained conversations. The model surpasses its competitors in both quality and cost, making it an ideal choice for businesses seeking efficient AI solutions without compromising performance.
Jamba-Instruct comes with safety features, chat capabilities, and improved command understanding out of the box, accelerating the time to production of enterprise applications and reducing total cost of ownership. AI21’s Senior Vice President, Pankaj Dugar, expressed excitement about the collaboration with Microsoft and the opportunity to introduce Jamba-Instruct to Azure customers, emphasizing their shared commitment to empowering businesses with innovative AI solutions.
Microsoft’s Corporate Vice President, Eric Boyd, also highlighted the collaboration, stating that Jamba-Instruct will soon be available in Microsoft Azure AI’s portfolio of models, offering developers a choice between open and basic models. With its unique architecture and context window length, Jamba-Instruct provides developers with a cost-effective option while maintaining exceptional quality and performance.
AI21, founded in 2017, creates core models and enterprise AI systems to accelerate the use of Generative AI in production. The company has raised $336 million from investors like NVIDIA, Intel, and Google and continues to offer consumer applications and enterprise solutions. Through the AI21 platform, companies can develop their own AI-based generative applications and services.
In conclusion, the collaboration between AI21 and Microsoft to integrate Jamba-Instruct into Azure AI marks an important milestone in bringing cutting-edge AI solutions to businesses. With its innovative architecture and performance capabilities, Jamba-Instruct offers Azure customers a transformative tool for language processing and efficiency, driving unprecedented levels of knowledge and productivity.
Article Source
https://www.prnewswire.com/news-releases/ai21-announces-collaboration-with-microsoft-to-bring-one-of-the-worlds-first-production-grade-mamba-based-models-to-microsoft-azure-ai-302151464.html