Qualcomm® AI Stack for developers and extension to on-device AI
summary
Qualcomm shows a unified Qualcomm AI Stack with ONNX Runtime support, tooling (Qualcomm Studio) and runtimes that compile models to Hexagon, CPU or GPU; a 1 billion-parameter text-to-image model producing 512×512 images runs on-device (an 8 Cx Gen 3 Snapdragon) and runs in ~5 W on Hexagon during demos. MLCommons April submissions back performance claims, an ultra-low-power Sensing Hub enables presence/auto-dim/privacy features, and a hybrid edge→cloud flow falls back when models exceed local capacity.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:40Generated AI
- 4:25generative AI
- 5:25ondevice AI
- 6:35evolution of AI
- 9:40use cases
- 13:40Alwayson Vision
- 15:00Benchmarks
- 16:30Qualcomm A
- 21:20Onyx Runtime
related talks
14mAccelerating On-Device AI Development
08mUnlocking AI at Scale for PyTorch Developers
15mBuilding Gen-AI experiences using LLM and LVM
16mRealize your vision with a guide to building Windows on Snapdragon
54mHarnessing On-Device AI to Transform the Smartphone Experience
26mHow to use Qualcomm AI Hub
37mHow to Optimize, Validate and Deploy ML Models On Device
17mOverview of Qualcomm Snapdragon Neural Processing Engine