Open quicksln opened 1 month ago
This is an integration design problem that will be addressed by the ONNX GenAI library repository in the future. To avoid relying on extra packages that might eventually be deprecated, we are holding off on this update until one ONNX library can work seamlessly with CPU, CUDA, and DirectML interfaces, ensuring long-term compatibility
Sounds logical, especially since the ONNX library is planned to work with CUDA, CPU, and DML directly. Switching between these three is not optimal. I know the reason behind it, but still.
Do you know when this update is expected—Q1 2025 or rather Q4 2025?
@quicksln , I don't have a specific time frame, hopefully Q1 2025, as soon this is in place will update the connector.
Describe the bug Look like Microsoft.SemanticKernel.Connectors.Onnx is missing the NuGet package release for CUDA and DirectML.
The project file already contains condition build configuration for:
https://github.com/microsoft/semantic-kernel/blob/main/dotnet/src/Connectors/Connectors.Onnx/Connectors.Onnx.csproj