XQuant Report Troubleshoot Pytorch¶

model_compression_toolkit.xquant.pytorch.facade_xquant_report.xquant_report_troubleshoot_pytorch_experimental(float_model, quantized_model, repr_dataset, validation_dataset, xquant_config)¶

Generate an explainable quantization report, detect degraded layaers and judge degrade causes for a quantized Pytorch model.

Parameters:

• float_model (torch.nn.Module) – The original floating-point Pytorch model.

• quantized_model (torch.nn.Module) – The quantized Pytorch model.

• repr_dataset (Callable) – The representative dataset used during quantization.

• validation_dataset (Callable) – The validation dataset used for evaluation.

• xquant_config (XQuantConfig) – Configuration settings for explainable quantization.

Returns:

A dictionary containing the collected similarity metrics and report data. Dict[str, Any]: A dictionary containing the analyze degrade cause report for degraded layaers.

Return type:

Dict[str, Any]