![]()
Microsoft has announced the upcoming preview of Models as a Service (MaaS) in Azure AI, which includes pay-as-you-go (PayGo) inference APIs and hosted fine-tuning for Llama 2, a large language model. This initiative expands Microsoft's partnership with Meta, marking Llama 2 as the first in this category through MaaS in Azure AI Studio. MaaS aims to simplify the development process for Generative AI developers by offering easy access to Llama 2 via API. The PayGo inference APIs are billed based on the number of tokens used, significantly lowering the entry barrier to utilizing Llama 2. These APIs are designed to integrate seamlessly with popular LLM tools like prompt flow, Semantic Kernel, or LangChain.
A key aspect of MaaS is its fine-tuning capability, allowing developers to adapt Llama 2 to their specific domain or problem, thus enhancing the model's predictive accuracy at a reduced cost. This feature is essential for personalizing the AI model to individual needs. Moreover, Azure AI's approach to content safety is integral to the service, with built-in moderation in the Llama 2 inference APIs to ensure the responsible use of AI.
Microsoft's introduction of MaaS reflects its commitment to democratizing AI and making advanced AI technology more accessible and adaptable. The PayGo Inference APIs and hosted fine-tuning capabilities for models like Llama 2 are expected to significantly reduce barriers to adoption, empowering organizations, developers, and data scientists of various skill levels and sizes?.