ai-hub-models - AI Models Designed for Optimal Performance on Qualcomm Platforms

Qualcomm® AI Hub Models

The Qualcomm® AI Hub Models are a diverse collection of cutting-edge machine learning models specifically optimized for efficient performance and seamless deployment on Qualcomm devices. These models are carefully designed to maintain low latency and minimal memory usage, making them ideal for applications on mobile platforms.

Key Features

Variety of Models: The AI Hub Models encompass a range of AI capabilities including vision, speech, text, and generative AI. They are fine-tuned for deployment directly on devices, which empowers developers to utilize sophisticated AI functionalities on smartphones, tablets, and other Qualcomm-enabled devices.
Open-source Recipes: Developers have access to open-source recipes that guide them through the process of quantizing, optimizing, and deploying these models. This transparency and accessibility ensure that developers can tailor models to suit specific needs and hardware constraints.
Performance Metrics: Users can browse detailed performance metrics to understand how these models perform across different Qualcomm devices. This information helps developers make informed decisions regarding model deployment based on their specific performance requirements.
Hugging Face Integration: The models are available through Hugging Face, a popular platform in the AI community. This integration provides researchers and developers with easy access to the models for testing and further development.
Sample Applications: Sample applications are available to demonstrate on-device deployment of AI Hub models. These applications serve as practical guides for developers looking to understand the real-world application of these models.
Sign-up for Hosted Devices: Users can sign up to execute models on hosted Qualcomm devices, offering hands-on experience with these models in a real deployment scenario. This feature is particularly valuable for testing and development purposes without needing physical hardware.

Technical Specifications

Supported Operating Systems: The AI Hub Models support Linux (x86, ARM), Windows (x86, ARM via x86 Python), and MacOS (x86, ARM).
Supported Runtimes: These models run on TensorFlow Lite, Qualcomm AI Engine Direct, and ONNX, ensuring wide compatibility with existing AI development frameworks.
Device Deployment: Models are deployable on Android, Windows, and Linux devices. Supported compute units include CPU, GPU, and NPU, which encompass Qualcomm's advanced processing units like Hexagon DSP and HTP.
Precision Levels: Models operate at various precision levels, including FP16 for floating points and INT8/INT4 for integers, optimizing for both performance and resource efficiency.
Supported Chipsets: Chipsets like Snapdragon 8 Gen series and Snapdragon X Elite are among the supported models, ensuring that the models run efficiently on some of the most popular and powerful mobile processors available.

Getting Started

To start using Qualcomm® AI Hub Models, ensure Python 3.8 to 3.10 is installed. It is recommended to create a Python virtual environment for managing dependencies—using tools like miniconda or virtualenv. From there, the base package qai_hub_models can be installed using pip. Additional dependencies for specific models, such as YOLOv7, can be included using the appropriate extension in the installation command.

Running Demos

Models are equipped with CLI demos to validate their functionality both locally and on-device. Local demos use PyTorch, while on-device demos require sign-up to run on Qualcomm's hosted cloud devices. These demos not only show the practical capabilities of the models but also ensure their readiness for deployment.

Conclusion

The Qualcomm® AI Hub Models provide a robust and versatile toolset for developers aiming to leverage machine learning on Qualcomm devices. With support for a wide range of devices, comprehensive documentation, and integration with popular AI frameworks like Hugging Face, these models are positioned as a significant asset in modern mobile and embedded AI development.

For any further assistance or information regarding these models, Qualcomm provides support through [email protected], ensuring that developers have the resources and guidance they need to succeed in their AI projects.