我最终使用HardTanh与如下设置

for key, bounds in self.coefficient_bounds.items():
    min_value, max_value = bounds
    m = torch.nn.Hardtanh(min_value, max_value)
    # the input needs to be a tensor so we have to convert
    # custom_pauli_coefficients[key] to a tensor
    input = torch.tensor(custom_pauli_coefficients[key])
    custom_pauli_coefficients[key] = m(input)