Why make parameter of tvm atleast_1d

converte parameter from scalar to 1d_array in func _parse_param file tensorflow.py.

        if array_ndim == 0:
            new_array = np.empty([1], dtype=np_array.dtype)
            new_array[0] = np_array
            self._nodes[name] = [tvm.relay.const(new_array)]

we couldn’t differentiate the shape of parameter between scale and 1d, when the node is constant.

example:

in_data0 = tf.placeholder(tf.float32, (), name="in_data0")
in_data1 = tf.constant(2)
tf.math.subtract(in_data0, in_data1, name="subtract")

the result of tvm different from one of tensorflow.

I want to know the intention of coder.