Question about schedulers in paper and in code

[HiGal]

Hello! I just read the code along with the paper and found that the paper stated the following.

During image generation, unless additional stated, Unless mentioned otherwise, we use 100 steps of PLMS sampler with a scale 5.

But in code, you decided to use LMSDiscreteScheduler.

vae, unet, text_encoder, tokenizer, image_encoder, mapper, mapper_local, scheduler = pww_load_tools(
            "cuda:0",
            LMSDiscreteScheduler,
            diffusion_model_path=args.pretrained_model_name_or_path,
            mapper_model_path=args.global_mapper_path,
            mapper_local_model_path=args.local_mapper_path,
        )

I am just curious and hope to hear an answer from you )

[csyxwei]

Hi, sorry for the misunderstanding. It is a typo in the paper, and we will revise it. The sampler we used is LMSDiscreteScheduler.

[HiGal]

Thank you for your answer!