| ### C# ORT Inference using VITISAI EP |
|
|
| This document describes how to set up C# ORT inference using VITISAI EP, with the AMD provided Nuget packages. |
|
|
| #### Structure of the repository |
|
|
| - **models/**: Contains pre-trained AI models and their configurations for three different use cases: |
| - `open-ai-npu-clip\` - CLIP text and vision models for image-text understanding |
| - `whisper-small-en\` - Whisper speech recognition model for English audio processing |
| - `yolov8\` - YOLOv8 object detection model for computer vision tasks |
|
|
| - **src/**: Contains C# source code examples demonstrating ONNX Runtime inference: |
| - `open-ai-npu-clip\Program.cs` - Runs inference using the clip models as stand alone models |
| - `open-ai-npu-clip\ClipInference.cs` - Runs clip inference using both models. Latency measurements and looping enabled. |
| - `whisper-small-en\Program.cs` - Whisper model inference example |
| - `yolov8\Program.cs` - YOLOv8 model inference example |
|
|
| #### Prerequisites |
|
|
| 1. Download and unzip the file containing the Nuget packages provided by AMD |
| 2. Clone the hugging face repository containing the application code, models, and cache |
| 3. `cd` into the cloned repo |
| 4. Create a directory for the Nuget packages inside the repo |
| ```bash |
| mkdir nupkgs |
| ``` |
| 5. Copy the contents of the unzipped Nuget packages to the nupkgs directory |
|
|
| 6. If dotnet is not installed, install using the executable from https://builds.dotnet.microsoft.com/dotnet/Sdk/9.0.305/dotnet-sdk-9.0.305-win-x64.exe |
|
|
| ### Instructions to create a new project |
|
|
| 1. From inside the cloned repo, create a new console application |
| ``` |
| dotnet.exe new console -n whisper_project |
| ``` |
| a new directory named `whisper_project` will be created |
|
|
| 2. Change into the project directory |
| ```bash |
| cd whisper_project |
| ``` |
| 3. Add the required Nuget packages to the project |
| ``` |
| dotnet add package System.Memory --version 4.6.3 |
| |
| dotnet add package System.Numerics.Tensors --version 9.0.9 |
| |
| dotnet add package Microsoft.ML.OnnxRuntime.Managed --version 1.23.0-dev-20250928-1204-d70213d40e --source path_to_nupkgs |
| |
| dotnet add package RyzenAI_Deployment --version 1.6.0 --source path_to_nupkgs |
| ``` |
| * The version numbers of the packages may change, please check the nupkgs directory for the correct version numbers. |
| * The path_to_nupkgs should be replaced with the actual path to the nupkgs directory created in step |
| * `whisper_project` is just an example |
|
|
| ### Instructions to run the provided examples |
|
|
| 1. Modify the `Program.cs` source code file in the `src\whisper-small-en` directory, replacing the paths that reference models and cache locations to the actual paths on your system. |
|
|
|
|
| 2. Copy the source code file from the `src` directory to the project directory created in the previous steps. |
| ``` |
| Copy ..\src\whisper-small-en\Program.cs . |
| ``` |
|
|
| 3. Build the project |
| ``` |
| dotnet build |
| ``` |
| 4. Run the project |
| ``` |
| dotnet run |
| ``` |
| * To run another example, repeat `Instructions to create a new project` and `Instructions to run the provided examples` steps with different project names and source code files. |
|
|
|
|
|
|
