Post training static quantization calibration inference

I’m trying to quantize a pretrained model which I haven’t its dataset for.
In the Quantization Workflow I see the following 5 step:

Calibrate the model by running inference against a calibration dataset

And again in the static quantization tutorial is used Dataloader with real data.

So the question is :
can I quantize a pretrained model without its dataset? (with just a random data of needed shape like in the torch.jit.trace())