quark.onnx.optimizations.convert_transforms_pipeline

Module Contents

Classes

class quark.onnx.optimizations.convert_transforms_pipeline.ConvertQDQToQOPTransformsPipeline(configs: Optional[Dict[str, Any]] = None)

Convert QDQ to QOperator transformations pipeline.

apply(model: onnx.ModelProto, candidate_nodes: Any, node_metadata: Any) → Tuple[onnx.ModelProto, Any]

Implement the transforms.

Parameters:

model – Onnx model to be quantized.

Returns:

Conveted onnx model.

class quark.onnx.optimizations.convert_transforms_pipeline.RemoveQDQTransformsPipeline(configs: Optional[Dict[str, Any]] = None)

Remove QDQ pairs transformations pipeline.

apply(model: onnx.ModelProto, candidate_nodes: Any, node_metadata: Any) → Tuple[onnx.ModelProto, Any]

Implement the transforms.

Parameters:

model – Onnx model to be quantized.

Returns:

Conveted onnx model.