export_config

`orchard.core.config.export_config` ¶

Export Configuration Schema.

Pydantic v2 schema defining model export parameters for ONNX. Supports optimization and validation settings.

Bases: BaseModel

Model export configuration for production deployment.

Defines ONNX export settings, optimization level, and validation parameters.

Attributes:

Name	Type	Description
`format`	`Literal['onnx']`	Export format (only 'onnx' supported).
`opset_version`	`PositiveInt`	ONNX opset version (18 recommended).
`dynamic_axes`	`bool`	Enable dynamic batch size for flexible inference.
`do_constant_folding`	`bool`	Optimize constant operations during export.
`quantize`	`bool`	Enable post-training quantization.
`quantization_type`	`Literal['int8', 'uint8', 'int4', 'uint4']`	Weight type — int8/uint8 for server, int4/uint4 for edge.
`quantization_backend`	`Literal['qnnpack', 'fbgemm']`	Backend — qnnpack for ARM/mobile, fbgemm for x86.
`validate_export`	`bool`	Validate exported model matches PyTorch output.
`validation_samples`	`PositiveInt`	Number of samples for export validation.
`max_deviation`	`PositiveFloat`	Maximum allowed output deviation for validation.
`benchmark`	`bool`	Run inference latency benchmark after export.

Example

cfg = ExportConfig(format="onnx", opset_version=18)