Untether AI is making waves in AI model deployment with its latest advancement—an innovative generative compiler that significantly expands model support and accelerates implementation.
Breaking Barriers in AI Model Support
The upcoming release of the imAIgine® Software Development Kit (SDK) will now support four times the number of AI models compared to previous versions. This means developers can deploy a wider range of neural networks, including those designed for object detection, semantic segmentation, classification, and error detection.
For users working on custom neural networks, the generative compiler can automatically generate new kernels, drastically reducing development time from weeks to mere minutes. This breakthrough enhances developer efficiency and speeds up AI innovation.
Push-Button Deployment with Generative Compiler
Traditional kernel-based compilers require labor-intensive C/C++ coding for each layer in a neural network, making scaling a challenge. Untether AI’s new compiler eliminates this bottleneck by automatically transforming high-level neural network operations into optimized hardware-ready representations.
This technology enables AI models to be deployed on Untether AI’s speedAI® devices in seconds, maximizing throughput while minimizing latency. The result? A seamless, efficient process for deploying complex neural networks at unprecedented speeds.
Optimized Performance Across Over 300 Models
Right out of the box, the enhanced imAIgine SDK supports more than 300 different neural network models, covering a broad range of AI applications. Additionally, the compiler is designed to work alongside pre-existing, hand-optimized kernels, ensuring the highest possible performance.
Seamless Integration with AI Acceleration Hardware
The generative compiler has been optimized for Untether AI’s At-Memory Compute architecture, supporting over 1,400 RISC-V processors. This allows for a seamless blend of existing and newly generated kernels, ensuring AI models achieve peak efficiency.
Accelerating AI Innovation for Enterprises
With this next-generation compiler, Untether AI is pushing the boundaries of AI development, making it easier for enterprises to deploy sophisticated AI models without extensive manual optimization.
For companies focused on optimizing AI-driven automation in industrial settings, Locus Robotics’ AI-driven warehouse solutions show how intelligent automation is transforming industries.
Availability and Early Access
The new generative compiler will be featured in the imAIgine SDK version 25.04, set for release in early April 2025. Developers interested in accessing this technology early can sign up for the early access program.
