ExportSettings
Defines the configuration for generating a portable ONNX-embedded DLL from a ZeroTrain model.
Properties of ExportSettings
| Name | Type | Description |
| Namespace | string | Specifies the namespace for the generated DLL. This allows the compiled engine to integrate seamlessly into the consumer's codebase. |
| ClassName | string | Specifies the class name of the generated inference model within the DLL. |
| Version | string | Specifies the version of the generated artifacts. This value is applied to the assembly metadata. |
| FileBaseName | string | Specifies the output file name of the generated artifacts. If not provided, a default name will be used. |
| ExportTarget | enumExportTarget | Specifies the output format for the export operation. |
| Export_Format | enumExportFormat | Specifies the output packaging format for the export operation. |
| Certification_Export_Location | enumFileExportLocation | Includes a certification file that validates deterministic behavior and ONNX parity. |
| Help_Export_Location | enumFileExportLocation | Includes a help file with usage instructions and integration guidance for the exported artifact. |
| OnnXOperationSet | int | Controls the ONNX operator set version used during model export. This setting should not be modified unless you fully understand ONNX runtime compatibility and operator support. |
| Distillation_Mode | enumDistillation | Controls how feature semantics are handled during ONNX distillation. Must be selected based on whether full semantic definitions are available. |
| DistillationMaxIterations | int | Maximum number of input combinations evaluated during distillation. Higher values increase coverage and accuracy but also increase processing time. Acts as a safety limit to prevent excessive computation. |
| Distillation_Inspection_Approach | enumDistillationInspectionApproach | Defines whether the ONNX model is analyzed for semantic hints (e.g., Min/Max ranges) during distillation. |
| Input_Mode | enumInputMode | Defines how input values are interpreted and must match the mode used when the model was built.
Natural:
- Accepts human-readable numeric values (e.g., 3894.72).
- The engine applies the configured ScaleFactor internally.
- Use this mode for standard integrations and user-facing inputs.
PreScaled:
- Expects values already scaled to the model's numeric space (e.g., 3894720 for ScaleFactor=1000).
- No scaling or transformation is applied at runtime.
- Intended for advanced scenarios requiring full control over numeric precision.
Important:
- The InputMode must match the mode used during model creation/export.
- Mismatched modes will produce incorrect results (e.g., zero confidence or invalid decisions). |
| MCP_Tool_Definition_Export_Location | enumFileExportLocation | Specifies where the generated MCP tool definition should be exported. |
| MCP_Tool_Definition | McpToolDefinitionBase | Optional MCP tool identity information used during export generation. |
Syntax
"ExportSettings": {
"namespace": "MyCompany.AI",
"className": "RiskEngine",
"version": "1.0.0",
"FileBaseName": "MyCompany.RiskEngine"
"ExportTarget": "DLL"
"InputMode": "PreScaled"
}