quark.onnx.finetuning.train_torch.train_model

Module Contents

Classes

class quark.onnx.finetuning.train_torch.train_model.ModelOptimizer

Optimizes weight or its rounding mode for the quantized wrapper module

classmethod run(quant_module: torch.nn.Module, inp_data_quant: numpy.ndarray[Any, Any] | List[numpy.ndarray[Any, Any]], inp_data_float: numpy.ndarray[Any, Any] | List[numpy.ndarray[Any, Any]], out_data_float: numpy.ndarray[Any, Any] | List[numpy.ndarray[Any, Any]], params: quark.onnx.finetuning.train_torch.train_model_param.TrainParameters) → None

Run the optimization for the target module :param quant_module: Quantized wrapper module which consists of a compute module and a optional act module :param inp_data_quant: Quantized wrapper module’s input data from all dataset, single array or array list :param inp_data_float: Original float module’s input data from all dataset, single array or array list :param out_data_float: Original float module’s output data from all dataset, single array or array list :param params: Optimization parameters