Unused pre_norm argument in ResnetBlock2D class

[masaishi]

I've been exploring the ResnetBlock2D class within the diffusers library, specifically located at diffusers/src/diffusers/models/resnet.py, and I noticed something that might warrant a discussion or clarification.

The constructor of ResnetBlock2D has an argument pre_norm that is fixed to True:

class ResnetBlock2D(nn.Module):
    def __init__(self, ..., pre_norm=True, ...):
        super().__init__()
        self.pre_norm = True
        # Other initialization code

https://github.com/huggingface/diffusers/blob/9cc59ba0891c3ccbebfcbb6784f4dde2d5bd8556/src/diffusers/models/resnet.py#L259C9-L259C29

I think the code should becomes self.pre_norm = pre_norm, and test result didn't change even if I change code.
If this modification is correct, I would like to send a pull request, but I am not familiar with the structure of models, and there is a possibility that it has some meaning and is fixed, so I posted this because I wanted to know about it.

[sayakpaul]

Cc: @yiyixuxu

[sayakpaul]

I think it could make sense default pre_norm to True and do what you proposed.

[masaishi]

Thank you for commenting. I will make a pull request tomorrow, because it can run test.