feat: enable attention dispatch for huanyuan video

[DefTruth]

enable attention dispatch for huanyuan video, adapted from:

https://github.com/huggingface/diffusers/blob/ac5a1e28fc9cc233863bcfb2abb9eef6807f156f/src/diffusers/models/transformers/transformer_hunyuanimage.py#L46C7-L46C32

Example

import torch
from diffusers import HunyuanVideoPipeline
from diffusers.utils import export_to_video

pipe: HunyuanVideoPipeline = HunyuanVideoPipeline.from_pretrained(
    "hunyuanvideo-community/HunyuanVideo",
    torch_dtype=torch.bfloat16,
)

pipe.enable_model_cpu_offload()
pipe.vae.enable_tiling()

def run_pipe():
    prompt = "A cat walks on the grass, realistic"
    output = pipe(
        prompt,
        height=320,
        width=512,
        num_frames=61,
        num_inference_steps=30,
        generator=torch.Generator("cpu").manual_seed(0),
    ).frames[0]
    return output

video = run_pipe()
export_to_video(video, "huanyuan_video.mp4", fps=15)

w/o this PR	w/ this PR

@yiyixuxu pipelines

[yiyixuxu]

thanks!

[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.

[sayakpaul]

Thanks! It would be nice to have some explanation of the change of dims being introduced in this PR.

[sayakpaul]

What's happening here?

[DefTruth]

Since transpose is no longer needed, at this point seq_len is located in dim=1.

[DefTruth]

This preprocess is same as: https://github.com/huggingface/diffusers/blob/ac5a1e28fc9cc233863bcfb2abb9eef6807f156f/src/diffusers/models/transformers/transformer_hunyuanimage.py#L46C7-L46C32

[sayakpaul]

Thanks for your contributions!