Expanding the Azure AI Model Catalog Ecosystem

As we gear up for this year's Microsoft Build, we're excited to announce a series of significant updates to our Azure model catalog. From going general availability (GA) soon to introducing new partnerships and models, these updates will enhance how developers integrate and utilize Gen across various platforms. 

General Availability of Models as a Service 

We're thrilled to announce that Models as a Service will reach General Availability in the coming weeks. This milestone marks a new era for MaaS, offering enhanced stability, scalability, and a host of new features designed to empower developers. 

New Models Launching at Build 

Get ready to explore new frontiers in with the launch of brand-new models in our catalog: 

Nixtla:  TimeGEN-1 model is a state-of-the-art generative pre-trained transformer foundation model designed specifically for time series forecasting. It's a powerful tool that can produce accurate forecasts from historical data without the need for retraining for each specific task.  This model can democratize access to advanced predictive insights, assisting both individuals and or…. Read more about Nixtla TimeGEN-1 here. 

Core42 JAIS: JAIS 30B Chat is now available on Microsoft's Azure AI model catalog, giving developers and businesses access to cutting-edge capabilities in both Arabic and English. This is one step closer to making AI and LLMs more accessible to speakers across the world. By leveraging a custom-built vocabulary, a massive pretraining corpus, and several architectural innovations, JAIS demonstrates great performance and efficiency in a variety of tasks, such as summarization, translation, text generation, and information retrieval. Read more about Core42 JAIS here. 

Looking ahead, we are excited to introduce several promising models that are set to join our catalog soon from Gretel, Bria, AI21, NTT and Stability AI. These upcoming additions highlight our commitment to delivering a diverse and powerful range of AI capabilities. 


New Partnership with Hugging Face: We're enhancing our model offerings through a new partnership with Hugging Face. This collaboration will bring an even broader array of AI models to our catalog, significantly expanding the tools available to our users. More on HuggingFace partnership. 


Expansion to Third-Party Developer Platforms: To make building generative AI applications even easier, we are extending Models as a Service to include integration with third-party developer platforms such as Arize, ClearML, and Dataloop. This expansion will streamline the development process, making it simpler to deploy and manage AI applications.  


Partnership with HiddenLayer for Model Safety: Safety remains a top priority, which is why we are partnering with HiddenLayer to ensure all models on our platform can meet the highest standards of security and reliability. This collaboration will help safeguard against potential vulnerabilities and ensure a secure environment for deploying AI models. 


Introducing Embeddings Benchmarks 

We are excited to announce the release of Embeddings Benchmarks into the Model Benchmarks experience in AI Studio. This is an expansion of our existing Model Benchmarks experience to now support the display of benchmarks for embedding models via an updated user interface. Users can now choose between viewing LLM & SLM quality benchmarks or embedding model benchmarks via tabs in the UI. An embedding model in AI is a type of model that learns to represent textual or non-textual data as vectors of numerical values, also known as embeddings. Embeddings can capture semantic and syntactic similarities between words, sentences, or documents, and enable downstream tasks such as natural language processing, computer vision, or recommendation systems. With Embeddings Benchmarks, users can compare the quality of different embeddings models across different tasks and datasets via the updated experience. New models, tasks and datasets will continue to be onboarded over time. 


Coming Soon: Performance Benchmarks 

As the Model Benchmarks product continues to grow, we would like to announce that coming soon, we will include support for performance benchmarks of models in this product. Performance will be measured via metrics such as throughput and latency, across various configurations. The inclusion of performance benchmarks will enable users to find the optimal model for their use case, and we are excited to bring this to our users soon. Stay tuned for more information about this feature and its upcoming release. 


Conclusion: These updates to Azure AI model catalog are designed to empower developers and businesses to harness the power of AI more efficiently. As we continue to innovate and expand our offerings, we look forward to seeing how our users leverage these new capabilities to create transformative solutions. Join us at Microsoft Build to learn more about these exciting developments and how they can revolutionize your AI projects. 



This article was originally published by Microsoft's AI - Machine Learning Blog. You can find the original article here.