diffusers _flash_attention_3 in dispatch_attention_fn is not compatible with the latest flash-atten interface.

Describe the bug

[FA3] Don't return lse: https://github.com/Dao-AILab/flash-attention/commit/ed209409acedbb2379f870bbd03abce31a7a51b7

but in the current diffuser version, it is not updated. https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_dispatch.py#L608

when use fa3 backend, diffuser will raise error.

Reproduction

import torch
import numpy as np
from diffusers import WanPipeline, AutoencoderKLWan, WanTransformer3DModel, UniPCMultistepScheduler
from diffusers.utils import export_to_video, load_image

dtype = torch.bfloat16
device = "cuda"

model_id = "Wan-AI/Wan2.2-TI2V-5B-Diffusers"
vae = AutoencoderKLWan.from_pretrained(model_id, subfolder="vae", torch_dtype=torch.float32)
pipe = WanPipeline.from_pretrained(model_id, vae=vae, torch_dtype=dtype)
pipe.to(device)

height = 704
width = 1280
num_frames = 121
num_inference_steps = 50
guidance_scale = 5.0


prompt = "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage."
negative_prompt = "色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走"
with attention_backend("_flash_3"):
    output = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    height=height,
    width=width,
    num_frames=num_frames,
    guidance_scale=guidance_scale,
    num_inference_steps=num_inference_steps,
    ).frames[0]
export_to_video(output, "5bit2v_output.mp4", fps=24)

Logs

System Info

🤗 Diffusers version: 0.35.0.dev0
Platform: Linux-5.4.119-19.0009.28-x86_64-with-glibc2.35
Running on Google Colab?: No
Python version: 3.11.13
PyTorch version (GPU?): 2.7.1+cu128 (True)
Flax version (CPU?/GPU?/TPU?): not installed (NA)
Jax version: not installed
JaxLib version: not installed
Huggingface_hub version: 0.34.3
Transformers version: 4.52.4
Accelerate version: 1.8.1
PEFT version: 0.15.2
Bitsandbytes version: 0.46.0
Safetensors version: 0.5.3
xFormers version: not installed

Who can help?

No response

Jul 30 '25 12:07 hmzjwhmzjw

Hi, thanks for the info! I will update accordingly. Since FA3 is in beta for now, interface breaks are expected for a while... For now, I can specialize the implementation on our end to detect and pass the correct arguments, and handle the return values as expected, but we will have to wait for a stable release to provide any kind of guarantee that our implementation will work