Generative AI and vision transformer models are now venturing to the edge, bolstered by Nvidia’s Jetson Orin. This advancement holds the promise of delivering contextual awareness, robustness, and a wide range of edge AI applications that can benefit industries and mobile machinery. These applications include video search, defect detection, robot autonomous navigation, human-robot interactions, and beyond.
Generative AI is the solution to challenges that conventional AI models, particularly convolutional neural networks in computer vision, struggle with. While CNN models show potential, they often require extensive post-processing and have difficulty with unfamiliar test data. The FAN + Dino vision transformer offers a robust alternative, outperforming CNN models in scenarios like detecting small objects in low-light conditions, while using only a quarter of the target data. Generative AI also dramatically improves ease of use with human languages. The zero-shot transformer model OWL + CLIP excels at people detection, vocabulary, and classification. It allows users to specify areas of interest and objects for detection with a simple text probe edit, all without prior training.
Generative AI accelerates the process of finding valuable information through real-time text-to-image and image-to-image searches powered by a vision transformer. This capability is fundamental for various multi-modal applications, operating in a shared embedded space that bridges text, images, and video. The AI visual agent leverages data from the edge and its contextual understanding to interact effectively with the real world.
With widespread access to edge AI and robotics application development and Nvidia’s comprehensive AI software stack and ecosystem, Jetson Orin delivers unmatched performance, memory capacity, and development experiences for generative AI at the edge.
Generative AI is set to play a pivotal role in IoT services, particularly in applications related to machine data, video-to-text, and speech-to-text. However, these use cases present unique challenges, including the processing of heavy data at high speed, ensuring the privacy and secure handling of sensitive video and audio data, and managing the escalating GPU costs needed for AI-hungry applications. Edge hardware appears to be the solution to these challenges, enabling real-time performance, data privatization, pre-processing, and efficient workload distribution, ultimately reducing GPU costs.
In response to these emerging needs, Ostream introduces GenRunner, the cornerstone of private Generative AI applications at the edge. GenRunner empowers high-impact use cases like video-to-text, private GPT, and speech-to-text. It resides within a compact single-board computer that can be easily scaled into clusters, serving as an integrated kit for Generative AI. GenRunner simplifies integration, reduces complexity, and accelerates time-to-market, with Director software serving as a mission control center for cluster deployment and AI performance optimization.
The key components of GenRunner include:
- GenRunner Node: The central computing unit at the heart of Gen AI at the edge.
- GAPI (GenRunner API): The software API gateway for seamless access to Generative AI capabilities.
- Director: Your mission control for cluster deployment and fine-tuning AI applications to ensure optimal performance.
Over the past decade, two prominent edge hardware devices, Raspberry Pi and Jetson Nano, have played crucial roles in IoT and AI development. With GenRunner, Ostream aims to establish a reputation for embeddable Gen AI, inspiring developers and integrators to explore the potential of Generative AI at the edge. The era of private Generative AI at the edge has arrived, and GenRunner leads the way.
For more information or to order your Ostream Nvidia Orin Generative AI Integration kit, click here.