Model conversion and optimization
Bring a trained PyTorch or ONNX model to Workbench, convert it for LiteRT, ONNX Runtime, or Qualcomm AI Runtime, and quantize or fine-tune it for accuracy.
Qualcomm AI Hub
Freemium
Optimize, validate, and deploy models for Qualcomm devices with hosted profiling, a model library, and sample apps.
O que é Qualcomm AI Hub?
Qualcomm AI Hub helps developers prepare AI models for use on Qualcomm devices. With Workbench, submit a trained PyTorch or ONNX model for conversion to LiteRT, ONNX Runtime, or Qualcomm AI Runtime. Quantize or fine-tune the model, then run inference and profile it on more than 50 types of hosted Qualcomm devices.
The hub also provides a catalog of more than 300 optimized ML and generative AI models, sample apps with bundled models, and GenieX, an open-source toolkit for running generative AI models locally on NPU, GPU, or CPU. The four components can be used independently or together.
O que Qualcomm AI Hub pode fazer?
Hosted device profiling
Run inference and profile model performance on more than 50 types of hosted Qualcomm devices.
Optimized model catalog
Browse more than 300 optimized ML and generative AI models, compare performance metrics, and find model assets for Qualcomm devices. Model licenses vary.
Sample apps and GenieX
Customize sample app repositories with a model already bundled, or use GenieX, an open-source toolkit with a unified runtime for NPU, GPU, and CPU.
Casos de uso
“Prepare a trained model for a Qualcomm device”
Bring a PyTorch or ONNX model to Workbench for conversion to an on-device runtime and optimization.
“Profile inference before integration”
Run and profile a model on hosted Qualcomm device types to inspect its device-specific performance before using it in an app.
“Compare model candidates”
Use the model catalog’s performance metrics to compare optimized ML and generative AI models for a Qualcomm device.
“Prototype an on-device application”
Start with a sample app repository that includes a model, or use GenieX to run a compatible generative AI model locally on NPU, GPU, or CPU.
Perguntas frequentes
What does Qualcomm AI Hub include?
It brings together Workbench, Models, Apps, and GenieX for on-device AI development on Qualcomm devices. The four components can be used independently or together.
How do I start using Workbench?
Create a Qualcomm ID, install the qai-hub Python package, and configure it with the API token in Workbench account settings. Run qai-hub list-devices to verify setup.
Are Workbench and AI Hub models free?
Workbench is currently free to use. Some models supplied by model makers are for purchase under their licenses; models marked download now are free and open. Check each model page for its license.
What model inputs and runtimes does Workbench support?
Workbench can convert trained PyTorch or ONNX models for LiteRT, ONNX Runtime, or Qualcomm AI Runtime. It also supports quantization, fine-tuning, inference, and profiling on hosted Qualcomm devices.
Informações rápidas
- Product
- On-device AI development for Qualcomm devices
- Workbench inputs
- Trained PyTorch and ONNX models
- Hosted device coverage
- 50+ types of Qualcomm devices
- Model catalog
- 300+ optimized ML and generative AI models
- Workbench setup
- Qualcomm ID and an AI Hub API token
Análise de tráfego de Qualcomm AI Hub
Os dados de tráfego são apenas para referência.
- Classificação do domínio
- 51
