Is this mistake regarding of lpw_stable_diffusion_xl.py

[ichi08]

I am reading lpw_stable_diffusion_xl.py (https://github.com/huggingface/diffusers/blob/main/examples/community/lpw_stable_diffusion_xl.py) to understand how to process prompt which over token limit 77.

And I found strange part in line 401 in the file I show following part.

for j in range(len(weight_tensor)):
            if weight_tensor[j] != 1.0:
                token_embedding[j] = (
                    token_embedding[-1] + (token_embedding[j] - token_embedding[-1]) * weight_tensor[j]
                )

If this part implies connecting before token smoothly, we should use "token_embedding[j-1]", I think. But I am not confident because it is 3 months that I started to study SD model...

Dose anyone can review this part with me?