Post-Quantization of Unet

[halo34]

Hi,
I try to post-quantize the Unet:
backend = "qnnpack" if arm else 'x86' qconfig = get_default_qconfig(backend) qconfig_dict = {"": qconfig} prepare_costume_config = PrepareCustomConfig() prepare_costume_config.set_non_traceable_module_names(["ConvTranspose2d"]) prepared_model = prepare_fx(model, qconfig_dict, prepare_custom_config=prepare_costume_config, example_inputs=example_input) quantized_model = convert_fx(prepared_model)
But I got the following error:AssertionError: Per channel weight observer is not supported yet for ConvTranspose{n}d.

But with prepare_costume_config.set_non_traceable_module_names(["ConvTranspose2d"]) I tried not to quantize this layer. Does anyone know how I can still exclude it?

[KumoLiu]

A similar issue on PyTorch.
pytorch/pytorch#54816