qualcomm/ai-hub-models

Qualcomm® AI Hub Models is our collection of state-of-the-art machine learning models optimized for performance (latency, memory etc.) and ready to deploy on Qualcomm® devices.

Qualcomm® AI Hub Models – Quick Overview

What it is – A Python package that gives developers a catalog of ready‑to‑run machine‑learning models (vision, language, generative AI, etc.) that have already been optimised for Qualcomm hardware (Snapdragon CPUs, GPUs, NPUs, Hexagon DSP, etc.). The package ships a command‑line tool (qai‑hub‑models) and a Python API for browsing the catalog, downloading the right model files, and running end‑to‑end demos.

Why it matters – Deploying deep‑learning models on phones, tablets, laptops, or IoT devices usually requires a lot of work: converting to a runtime (TensorFlow Lite, ONNX, Qualcomm’s AI Engine), quantising, compiling for a specific chipset, and testing performance. Qualcomm’s AI Hub Models bundle all of that work so you can focus on the application rather than the low‑level optimisation.

Key Features (from the README)

Feature What you get
Huge model catalog Over 100 models covering image generation, text generation, and classic computer‑vision tasks. Each entry links to a model‑specific README with details and download links.
Device‑specific optimisation Models are pre‑quantised (e.g., w8a8), compiled for runtimes like LiteRT, ONNX, or Qualcomm AI Engine Direct, and tested on real devices (Galaxy S series, Xiaomi, Snapdragon X Elite, automotive platforms, etc.).
CLI & Python API qai‑hub‑models lets you browse (models), view info (info), see numerics (numerics), performance (perf), fetch assets, install dependencies, export, evaluate, and run demos – all from the terminal or within Python code.
One‑click export & evaluation With a Qualcomm AI Hub Workbench account you can quantise, compile, profile, and run inference on a cloud‑hosted Qualcomm device, then compare on‑device results with PyTorch reference outputs.
Sample apps Native Android/Linux/Windows sample apps are in the companion AI Hub Apps repo, and Python demo apps wrap preprocessing/post‑processing for each model.
Supported runtimes & hardware • Qualcomm AI Engine Direct (Android, Linux, Windows) • LiteRT (TensorFlow Lite) • ONNX (QNN execution provider) • CPU, GPU, NPU (Hexagon DSP, HTP) with FP32/FP16/INT8/INT16 precision.
Installation options Light‑weight CLI only (qai‑hub‑models‑cli) or full package (qai_hub_models) that includes export/evaluation capabilities.

Typical Workflow

  1. Configure access – Create a Qualcomm ID, log into AI Hub Workbench, and store your API token with qai‑hub configure.
  2. Pick a model – Browse the catalog (qai‑hub‑models models) and read the model’s README for supported runtimes and precision.
  3. Install dependencies – qai‑hub‑models install <model> pulls any extra Python packages needed for that model’s demo.
  4. Export – qai‑hub‑models export <model> --target-runtime tflite --precision w8a8 --device "Samsung Galaxy S25" quantises, compiles, and uploads the model to a cloud‑hosted device.
  5. Evaluate – qai‑hub‑models evaluate <model> --hub-model-id <ID> runs inference on the device, compares results to a PyTorch baseline, and reports accuracy/performance.
  6. Run a demo – Either locally with PyTorch (demo <model>) or on‑device (demo <model> --eval-mode on-device).

Who Should Use This

  • Mobile/IoT developers who need high‑performance AI on Snapdragon‑based phones, tablets, laptops, or embedded boards.
  • Researchers wanting to test a model on real hardware without writing their own compilation pipelines.
  • Product teams building AI‑enhanced apps (camera, AR, voice, translation) that must run offline on‑device.

Limitations & Gotchas

  • The full package requires a Qualcomm AI Hub Workbench account and internet access to the cloud‑hosted device for export/evaluation.
  • Windows ARM64 Python is not supported for Snapdragon X Elite/X2 Elite; use AMD‑64 Python.
  • Some demos are PyTorch‑only and do not run on‑device; check each model’s README for supported demo modes.

Bottom Line

Qualcomm AI Hub Models is a practical, vendor‑backed toolbox that turns a long list of state‑of‑the‑art ML models into ready‑to‑deploy assets for Qualcomm hardware, with a convenient CLI, performance data, and end‑to‑end demos.

Related

  • Project
    qualcomm/ai-hub-apps

    Qualcomm AI Hub Apps provides sample Android, Windows, and Ubuntu applications plus tutorials that demonstrate how to run vision, audio, and language models on Snapdragon devices using TensorFlow Lite, ONNX, and Qualcomm’s Genie SDK, with optional CPU/GPU/NPU acceleration.

  • Project
    Arm-Examples/ML-examples

    A collection of machine learning examples and tutorials for deploying and optimizing AI models on Arm Ethos NPUs and Cortex-based platforms.

  • Project
    amd/RyzenAI-SW

    A suite of tools and runtime libraries for optimizing and deploying AI inference on AMD Ryzen AI-based PCs to maximize hardware acceleration.

  • Project
    google-ai-edge/litert-samples

    Google AI Edge LiteRT Samples is a genuine open‑source collection of on‑device machine‑learning demos, model‑conversion recipes, and automation tools for LiteRT (the successor to TensorFlow Lite) and LiteRT‑LM. It provides Android, iOS, and Web/WASM examples covering speech, vision, text‑to‑image, and multimodal apps, plus scripts to convert Hugging Face checkpoints into LiteRT‑compatible bundles. The repo is aimed at developers building high‑performance edge AI applications.

  • Project
    OpenSQZ/MiniCPM-V-CookBook

    A comprehensive implementation guide and deployment cookbook for the MiniCPM series, enabling the integration of vision, speech, and text AI capabilities across servers and edge devices.