Potential mistake in positional encoding example

Hello,
The example notebook for creating a Transformer network features a section on how to add a positional encoding to the input sequence.

As basis, the formula from Attention Is All You Need is given:

$\Large{PE_{(pos, 2i)} = \sin(pos / 10000^{2i / d_{model} })}$
$\Large{PE_{(pos, 2i+1)} = \cos(pos / 10000^{2i / d_{model} })}$

To my understanding, the here described PE does alternate between sine and cosine functions for even and odd feature dimensions.

The implementation in the tutorial however calculates sine functions for the first half of feature dimensions and cosine functions for the second half:

def positional_encoding(length, depth):
    depth = depth/2

    positions = np.arange(length)[:, np.newaxis]     # (seq, 1)
    depths = np.arange(depth)[np.newaxis, :]/depth   # (1, depth)

    angle_rates = 1 / (10000**depths)         # (1, depth)
    angle_rads = positions * angle_rates      # (pos, depth)

    pos_encoding = np.concatenate(
        [np.sin(angle_rads), np.cos(angle_rads)],
        axis=-1
    ) 

    return tf.cast(pos_encoding, dtype=tf.float32)

Am I missing something here or is this implementation different from the mathematical definition?

Open docs/tutorials/transformer.ipynb and inspect the positional_encoding function alongside the displayed Attention Is All You Need formulas. Verify whether the sine/cosine ordering is intentional or inconsistent; done means resolving the reported mismatch in the example or its explanation.