Qualcomm and ONNX: Pioneering Generative AI on Edge Devices


Qualcomm and ONNX: In the rapidly evolving landscape of artificial intelligence (AI), portability and efficiency are key. ONNX (Open Neural Network Exchange), an open format for representing neural networks, was developed by Microsoft and Facebook in 2017 to address these needs. Qualcomm, a leading name in the tech industry, has embraced ONNX to bring generative AI models to edge devices through its Snapdragon processors. This article delves into how Qualcomm’s support for ONNX is revolutionizing the capabilities of generative AI on mobile and other edge devices.

What is ONNX?

ONNX is an open-source format designed to make neural networks portable across various frameworks and platforms. This enables developers to deploy their AI models more easily and efficiently. Qualcomm has been a strong advocate for ONNX, integrating its support into the Qualcomm Neural Processing Engine (NPE) SDK, which is compatible with Snapdragon processors.

Qualcomm and Generative AI

Generative AI is a subset of AI that focuses on creating new content—be it images, text, or music—from scratch. Qualcomm is leveraging its ONNX support to enable a range of generative AI applications on mobile devices. These applications span multiple domains, including augmented reality (AR), virtual reality (VR), and machine translation.

Notable Generative AI Models on Snapdragon

Qualcomm has showcased several generative AI models that run efficiently on Snapdragon processors:

  • ControlNet: A 1.5 billion parameter image-to-image model for generating images with precise control.
  • Whisper: A 1.6 billion parameter Automatic Speech Recognition (ASR) model capable of transcribing and translating audio in multiple languages.
  • Bloom: A 1.5 billion parameter language model for generating text, translating languages, and crafting creative content.
  • chatGLM: A 7 billion parameter language model designed to answer questions comprehensively and informatively.
  • GPT-J: A 6 billion parameter language model for text generation, language translation, and creative writing.

These models are optimized to run on a variety of edge devices, including smartphones, PCs, XR headsets, vehicles, and IoT devices.

Advantages of Edge Computing for Generative AI

Running generative AI models on edge devices offers several benefits:

  • Reduced Latency: No need to send data to the cloud, ensuring quicker response times.
  • Increased Privacy: Data stays on the device, safeguarding user privacy.
  • Cost-Efficiency: Eliminates the need for a cloud connection, reducing operational costs.
  • Improved Reliability: Less dependency on network connectivity enhances overall reliability.

Real-World Applications

Here are some practical use-cases for Qualcomm ONNX and generative AI:

  • A photographer can utilize a ControlNet model on their smartphone to create images with specific styles.
  • A traveler can employ a Whisper model on their laptop for real-time audio transcription and translation.
  • A student can use a Bloom model on their tablet to summarize complex topics or write creative essays.
  • A customer service representative can leverage a chatGLM model to answer customer queries effectively.
  • Developers can use a GPT-J model to auto-generate code, translate languages, or create diverse types of content.


Qualcomm’s commitment to ONNX and generative AI is making it easier and more efficient to run advanced AI models on a wide range of devices. As generative AI continues to evolve, Qualcomm’s technologies are set to play a pivotal role in making these advancements accessible to the masses.

For further information, you can visit Qualcomm’s official website and ONNX’s GitHub repository.

Leave a Reply