The deployment of AI models has long been a complex and challenging task for developers. While building AI-powered applications has become easier with modern tools, hosting the models and taking care of the infrastructure has remained a headache for many. Microsoft aims to address this issue by introducing Models-as-a-Service (MaaS), a solution that simplifies the deployment process and allows developers to focus more on the creative aspects of their work.
MaaS is the AI equivalent of cloud services, where developers can access pre-trained models without the need for virtual machines. With MaaS, developers can rent inference APIs and host fine-tuning through a pay-as-you-go plan, making it more convenient and cost-effective. Microsoft already offers over 1,600 models for various purposes, and MaaS allows developers to easily leverage the AI functionality into their software.
Seth Juarez, the principal program manager for Microsoft’s AI platform, highlights the complexity of deploying AI models. He describes the process as a series of combinations of incantations, Pytorch versions, CPU and GPU configurations, which can be frustrating and time-consuming. MaaS abstracts these complexities, providing developers with a catalog of models that they can easily access and deploy as endpoints with just a button click.
Since its inception in 2023, Microsoft has made select models available through MaaS. Initially, Mistral-7B and Meta’s Llama 2 were offered, and recently, TimeGen-1 from Nixtila and Core42 JAIS have been added. More models from AI21, Bria AI, Gretel Labs, NTT Data, Stability AI, and Cohere are also expected to be available soon. However, not all models can be included in MaaS due to their specialized nature, and developers may need to deploy them differently.
Juarez envisions a future where developers will have the option to “rent” or “own” their models, similar to being homeowners or renters. In the “rent” scenario, developers can leverage MaaS, letting Microsoft handle the upkeep and maintenance of the models. On the other hand, in the “own” scenario, developers have more control over the deployment and management of their models, but also need to take care of the infrastructure.
The rise of AI as a prominent technology can be compared to the popularity of cloud computing. Juarez suggests that the demand for AI features and services has reversed the traditional model of tech companies pushing out new technologies. Instead, users are now demanding specific AI capabilities, driving the research and commercialization of AI. This trend is evident in the increased usage of AI-based applications like ChatGPT by consumers, which is now fueling the enterprise sector to catch up with user demands.
In conclusion, Microsoft’s Models-as-a-Service (MaaS) offers developers a simplified and convenient way to deploy AI models. By abstracting the complexities of hosting and infrastructure management, MaaS enables developers to focus on the creative aspects of their work. With a growing catalog of models and the option to rent or own them, MaaS is poised to empower developers and accelerate the adoption of AI technologies across various industries.
Source link
