Problem Statement
Currently, users can configure ONNX execution providers only by manually creating a genai_config.json file next to the ONNX model. This approach is cumbersome and doesn't integrate well with the Semantic Kernel configuration patterns.
Proposed Solution
Add programmatic support for provider configuration in the ONNX connector, allowing users to specify execution providers directly through the API instead of requiring manual JSON configuration files.
Benefits
- Better Developer Experience: Configure providers through code instead of manual file editing
- Integration: Aligns with Semantic Kernel's configuration patterns
- Flexibility: Dynamic provider selection based on runtime conditions
- Maintainability: Eliminates the need for external configuration files
Breaking Changes
None - this would be a purely additive feature maintaining backward compatibility. The existing genai_config.json approach would continue to work.
NB: I will suggest an implementation.
Problem Statement
Currently, users can configure ONNX execution providers only by manually creating a
genai_config.jsonfile next to the ONNX model. This approach is cumbersome and doesn't integrate well with the Semantic Kernel configuration patterns.Proposed Solution
Add programmatic support for provider configuration in the ONNX connector, allowing users to specify execution providers directly through the API instead of requiring manual JSON configuration files.
Benefits
Breaking Changes
None - this would be a purely additive feature maintaining backward compatibility. The existing
genai_config.jsonapproach would continue to work.NB: I will suggest an implementation.