[IF| add set_begin_index for all IF pipelines

[yiyixuxu]

IF use the "quaredcos_cap_v2" beta scheduler, it is likely to see very large k-sigmas at initial steps, which translate to the initial duplicated timesteps.

I added set_begin_index for IF pipelines since we have already implemented for this exact problem in img2img and inpainting pipelines and it is very easy to extend this to text-to-image; This should fix #7214

import torch
import gc
from diffusers import DiffusionPipeline
from diffusers import DPMSolverMultistepScheduler


pipe = DiffusionPipeline.from_pretrained(
    f"DeepFloyd/IF-I-XL-v1.0",       
    variant= "fp16",
    torch_dtype= torch.float16,
    requires_safety_checker= False,
    feature_extractor= None,
    safety_checker= None,
    watermarker= None,
)
       

pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config, algorithm_type='sde-dpmsolver++', use_karras_sigmas=True)

positivePrompt="an americana painting of a storm over the colorado mountains"
negativePrompt="blurry"

positivePrompt="an americana painting of a storm over the colorado mountains"
negativePrompt="blurry"

with torch.no_grad():
    positive, negative = pipe.encode_prompt(
        prompt=positivePrompt,
        negative_prompt=negativePrompt,
    )
pipe.text_encoder = None
gc.collect()
torch.cuda.empty_cache()

pipe.enable_model_cpu_offload()

images = pipe(
    prompt_embeds=positive,
    negative_prompt_embeds=negative,
    height=64,
    width=64,
    generator=torch.manual_seed(42),
    guidance_scale=7,
    num_images_per_prompt=1,
    num_inference_steps=20,
    output_type="pt",
).images

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.