Question about RoPE code

[rangehow]

I found here exist a difference in rope implementation mostly on permutation. Does this difference not affect the final result ？ I'm not quite sure what I'm thinking. Sincerely ask for your advice : )

Paper version should be:

version in this repo:

[vpj]

The ordering is different. So it wont affect training from scratch but you cant load a model trained with different ordering.

[rangehow]

Thanks for your answer : ) Is there exist some reason that the latter implementation was widely used in code instead former one ?

[vpj]

It's easier to code