Microsoft Azure is increasingly becoming the preferred cloud platform for organizations. It leads the way by providing tools and services to develop and deploy artificial intelligence-based applications.
As organizations’ needs grow, they require sophisticated AI-based solutions to achieve their business goals. To support the development and deployment of AI applications, Azure AI provides flexible and advanced tools and models to meet the unique and complex needs of today’s organizations.
Microsoft has introduced several updates to help application developers quickly create AI solutions using the Azure AI platform.
Let us discover the improvements in Azure to make AI development easier and quicker.
- Improvements to the Phi family of models (AI models developed for natural language processing), including the introduction of a new Mixture of Experts (MoE) model and language support for 20+ languages.
- The availability of the Jamba 1.5 family of open models in Azure AI model as a service
- Integrated vectorization for Azure AI search with data prep and embedding
- Use of generative AI to extract custom fields from documents with accuracy
- Availability of Text to Speech Avatar
- General availability of VS code extension for Azure Machine learning
- Conversational PII Detection Service in Azure AI language
Phi model family with support for more languages
Among the new models introduced in the Phi family is Phi-3.5-MoE, a Mixture of Experts (MoE) model. This model combines 16 smaller experts to provide users the speed and computational efficiency.
While the model has 42 billion parameters, it utilizes 6.6 billion active parameters at any one time. The MoE model trains these experts, or subsets of parameters, for specific tasks. By specializing experts during training and using the relevant experts for specific tasks, the MoE model can handle complex tasks more efficiently, improving computational performance.
Another addition is the Phi-3.5-mini model. Both the MoE model and the mini model support over 20 languages. This multilingual support allows users to interact with the models in the language they are most comfortable with.
AI21 Jamba 1.5 Large and Jamba 1.5 on Azure AI model as a service
The availability of Jamba OpenAI models, such as Jamba 1.5 Large and Jamba 1.5, in Azure AI is another update. These open-source large AI models, based on Jamba’s Mamba Transformer MoE architecture, support innovative technologies to enhance AI performance. They are particularly suited for handling complex use cases, such as processing lengthy documents and long-context scenarios, and other complex use cases, making them ideal for industries like finance, healthcare, and retail.
Simplify RAG
Microsoft simplifies Retrieval-Augmented Generation (RAG) pipelines by integrating comprehensive, end-to-end data preparation and embedding capabilities. This enables organizations to leverage RAG in generative AI applications, allowing them to tap into their private data without requiring model retraining.
RAG enables users to access advanced retrieval strategies like vector and hybrid retrieval to produce useful and context-specific information for a given query. However, the introduction of integrated vectorization in Azure AI search simplifies the search process. It automates the various processes of data preparation, making it easier and more efficient to work with data living in multiple sources.
In addition to enhancing developer efficiency, integration vectorization enables organizations to provide complete RAG solutions for new projects, allowing teams to quickly create custom applications tailored to their specific datasets.
Extract custom fields in Document Intelligence
Now users can extract custom fields from unstructured documents by building and training a custom generative AI model within document intelligence. This capability uses generative AI to pull put user specific fields from various document types.
Generate engaging experiences with custom avatars
Text to Speech Avatar within Azure AI speech service brings natural-sounding voices and phot realistic avatars, across various languages and voices. This enhances customer experience and engagement. TTS provides various pre-built avatars that comes with a range of natural-sounding as well as provides users with the capability to create custom voices using Azure Custom Neural Voice. The photorealistic avatars can be adapted to match the company’s branding.
VS Code extension for Azure Machine Learning
VS Code extension for Azure Machine Learning allows users to create, train and deploy machine learning models from VS Code environment, whether on desktop or web. This helps organizations to streamline workflow, enhance collaboration, and productivity.
Conversational PII Detection Service
General availability of Conversational PII Detection Service in Azure AI Language enhances Azure’s AI ability to identify and edit sensitive information in conversations. This helps to improve data privacy and security for developers creating generative AI apps for their organizations.
With the announcement of new capabilities, organizations can create smart AI-powered applications that enhance customer experiences and attract more customers to their business.
To know more about Azure OpenAI and the various other innovations to help you build, deploy, and scale your AI solutions, you can connect with IAX DYNAMICS, the leading Microsoft Azure Partner.
We are a Microsoft Gold Partner with proficiency in offering Microsoft Cloud Services to organizations in various sectors. Through management solutions for Azure OpenAI Service, our Azure professionals guide you to leverage this smart technology to your business advantage.
FAQ SECTION
What is the significance of the Phi-3.5-MoE model in Azure AI?
The Phi-3.5-MoE model is a Mixture of Experts (MoE) model that combines 16 smaller experts to provide speed and computational efficiency, making it ideal for complex tasks.
How does integrated vectorization in Azure AI search simplify the search process?
Integrated vectorization automates data preparation processes, making it easier and more efficient to work with data from multiple sources, and enables organizations to provide complete RAG solutions for new projects.
What is the benefit of using generative AI to extract custom fields from documents in Document Intelligence?
A: Generative AI enables users to extract custom fields from unstructured documents with high accuracy, allowing organizations to use their private data without requiring model retraining.
What is the purpose of the Text to Speech Avatar in Azure AI speech service?
The Text to Speech Avatar provides natural-sounding voices and avatars that resemble images, enhancing customer experience and engagement, and allowing users to create custom voices and adapt avatars to match their company’s branding.
