question about rotary positional encoding (rope) for text.

hi authors, thanks a lot for releasing the code!

i noticed that position ids for text embeddings is set to zeros at this line: https://github.com/black-forest-labs/flux/blob/c23ae247225daba30fbd56058d247cc1b1fc20a3/src/flux/sampling.py#L51

i am wondering why this is the case, esp. when compared to the image position ids being a 2D meshgrid here: https://github.com/black-forest-labs/flux/blob/c23ae247225daba30fbd56058d247cc1b1fc20a3/src/flux/sampling.py#L42

is it because t5 model already adds positional encoding to the text, perhaps?

thanks!