Qualcomm AI Hub explained: Workbench, Models and Apps
Sign up for Developer monthly newsletter
Join thousands of developers around the globe who receive latest news and updates from our monthly curated newsletter.
Sign upQualcomm AI Hub is where you go to accelerate your on-device AI model development. It is our one-stop-shop for enabling AI models to be deployed across edge devices through three offerings: an on-device model optimization platform, pre-optimized models, and sample applications.
To better distinguish our products, our on-device model optimization platform, previously called Qualcomm AI Hub, is now Qualcomm AI Hub Workbench.
It’s the same platform you know, with a name that reflects what it stands for: a hands-on, developer friendly platform where you go to hammer on your trained model and bang it into shape so that it is ready to deploy on device. We’re excited to see all the great work you’ll accomplish using Qualcomm AI Hub.
What is Qualcomm AI Hub?
Qualcomm AI Hub is the overarching name for all three products: Qualcomm AI Hub Workbench, Qualcomm AI Hub Models, and Qualcomm AI Hub Apps. Developers can use each product independently or together to accelerate the development of their on-device AI models.
Qualcomm AI Hub Workbench is the platform for on-device AI. You can bring any model, target any device, and use any runtime to have an optimized, deployable asset within minutes.
Qualcomm AI Hub Models are our pre-optimized models to select your model, review performance metrics across various devices with Qualcomm technology, and easily download model assets.
Qualcomm AI Hub Apps are a set of sample apps for learning how to integrate your model into your application and achieve optimized performance on-device.
Now, let's dig into when you should use each product and their differences:
Qualcomm AI Hub Workbench is our powerhouse platform, focused on bringing your custom trained model on device by leveraging simple-to-use Python APIs. This is where you optimize your model, quantize and profile it, running inference on real devices, validating that it meets your performance requirements.
We focus on doing the heavy lifting behind the scenes – automatic optimizations, compiling for your desired runtime, visualizing which layers are bottlenecks – so that you don’t have to.
- Compile your model for your desired runtime and device.
- Quantize at various precisions to review performance.
- Profile your compiled model on a real device, to run inference and collect detailed performance measurements.
- Inference on-device with your own input data, check accuracy and review model outputs.
Our comprehensive API documentation shares what additional options you have for each job.
Qualcomm AI Hub Models contains 250 pre-optimized models (and growing!) that encompasses popular use cases, from object detection to speech recognition, video classification, image to text, and more. We run every model from Qualcomm AI Hub Models across a variety of chipsets used in mobile, compute, IoT, and auto devices.
Every two weeks, these models are compiled and profiled on Qualcomm AI Hub Workbench, running inference to provide up-to-date performance metrics across applicable runtimes.
These models are ready to use: simply select your model and download to integrate into applications. The model cards also serve as a point of reference for which models are compatible on certain chipsets.
Once you’ve selected one of our Qualcomm AI Hub Models, check out each model’s specific instructions in the corresponding model folder in our GitHub Repository (quic/ai-hub-models). From there, you can run an end-to-end sample, explore a PyTorch-based demo, and generate the optimized model yourself via Qualcomm AI Hub Workbench, running inference and downloading the model.
Qualcomm AI Hub Apps is a new addition to Qualcomm AI Hub, consisting of sample applications to help you learn how to integrate our Qualcomm AI Hub Models into your application. Walk through instructions per application, in our GitHub repository (quic/ai-hub-apps) for Windows and Android.
You can leverage our samples and learn how to configure your runtime and match performance results from Qualcomm AI Hub Models. Drag and drop your selected model into the specified location, set up the application and away you go, ready to deploy on device!
Where real work gets done to bring your model on device
All three products work together harmoniously to ensure that you’re bringing the best version of your model on-device. Explore and select from Qualcomm AI Hub Models, build and experiment on model performance in Qualcomm AI Hub Workbench, and bundle your model into an application with Qualcomm AI Hub Apps to deploy on device. We can’t wait to see what on-device AI experiences Qualcomm AI Hub enables you to build.
To dive in, create an account on Qualcomm AI Hub Workbench.
Join our community on Slack to ask any questions about Qualcomm AI Hub.
Come for support, stay for the community
Get support from experts, connect with like-minded developers, and access exclusive virtual events.
