quark.torch.quantization.graph.processor.processor

Module Contents

Functions

quark.torch.quantization.graph.processor.processor.transform_for_annotation(model: torch.fx.GraphModule) → torch.fx.GraphModule

Prepare before annotation, for both PTQ and QAT

quark.torch.quantization.graph.processor.processor.freeze_model(model: torch.fx.GraphModule) → torch.fx.GraphModule

After quantization, we need to export model (e.g onnx, torch.export), we regard the users will not need further calibration, training, optimization.