Absolute Position Encoding：Why are the two tensors not alternately merged？

[davinca]

tensor2tensor/tensor2tensor/layers/common_attention.py

Line 453 in bafdc1b

signal = tf.concat([tf.sin(scaled_time), tf.cos(scaled_time)], axis=1)

In the orginal paper, the position_embedding is like this: [..., sin i, cos i, ...]

[martinpopel]

See #177 and #1591 (and #1677).

[davinca]

just different orderings of the same set of channels, The effects of both are consistent theoretically.