Obtaining models

There are two main artifact types for DENKflow users.

.denkflow: Complete exported pipelines

A .denkflow file contains the full pipeline configuration and the export target selected in the Vision AI Hub.

Use .denkflow when you want:

• the easiest deployment path
• preconfigured preprocessing and postprocessing
• quantized exports such as Intel CPU/GPU/NPU, DirectML QDQ, or TensorRT INT8
• less custom runtime wiring in your application

How to get a .denkflow file

1. Train or select a network in the Vision AI Hub.
2. Request or confirm a valid license.
3. Open the model details page.
4. Open the Exports tab.
5. Click Add New Model Export.
6. Choose the export target that matches your deployment hardware.
7. Wait for export creation.
8. Download the generated .denkflow file.

Typical export targets

Examples include:

• CPU_FP32_ONNX
• CUDA
• INTEL_CPU
• INTEL_GPU
• INTEL_NPU
• DIRECTML_FP32_ONNX
• DIRECTML_INT8_QDQ_ONNX
• INT8_NVIDIA_GPU_TENSORRT
• INT8_NVIDIA_JETSON_TENSORRT

Pick the export target based on your deployment machine, not on your development workstation.

.denkmodel: Individual models for custom pipelines

A .denkmodel file is the right choice when you want to build your own pipeline graph.

Use .denkmodel when you want:

• custom routing between nodes
• custom pre- or post-processing logic
• manual runtime selection in code
• multiple models in a hand-built pipeline

How to get a .denkmodel file

1. Open the network details page in the Vision AI Hub.
2. Use Download Network.
3. Save the .denkmodel artifact locally.

Important difference

• .denkflow can represent quantized export targets
• .denkmodel downloads are currently intended for custom pipeline assembly and are not the main path for Hub-driven quantized deployment

If you are unsure which artifact to choose, use .denkflow.