Fix padding in dreambooth

[shirayu]

Fix padding in dreambooth as it is done in inference.

diffusers/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py

Lines 231 to 236 in de00c63

	text_inputs = self.tokenizer(
	prompt,
	padding="max_length",
	max_length=self.tokenizer.model_max_length,
	return_tensors="pt",
	)

• Reference by kohya_ss (in Japanese):
  • They say this is one of the difference between https://github.com/XavierXiao/Dreambooth-Stable-Diffusion/ and https://github.com/huggingface/diffusers and padding like this PR made better results

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[shirayu]

I do not know which is theoretically correct, before or after the modification. I am looking for other opinions.

[patrickvonplaten]

@patil-suraj could you take a look here?

[patil-suraj]

Thanks for the PR, lgtm! Not sure if this improves the but it's okay to do this as that's how the model was trained.