MAC Ultra pipe.load_lora raise an error "AttributeError: QwenImageDiT has no attribute `diffusion_model`"

[donaldxu]

diffsynth-engine==0.6.0

source code

import gradio as gr
from diffsynth_engine import fetch_model, QwenImagePipeline, QwenImagePipelineConfig

import torch

model_name = "Qwen/Qwen-Image-2512/"
device = "mps" # Mac 使用 MPS (Metal Performance Shaders)
torch_dtype = torch.bfloat16

print("Loading model...")

transformer_files = [f'Qwen/Qwen-Image-2512/transformer/diffusion_pytorch_model-0000{x}-of-00009.safetensors' for x in range(1,10)]
encoder_files = [f'Qwen/Qwen-Image-2512/text_encoder/model-0000{x}-of-00004.safetensors' for x in range(1,5)]
vae_files = [f'Qwen/Qwen-Image-2512/vae/diffusion_pytorch_model.safetensors']
config = QwenImagePipelineConfig.basic_config(
model_path = transformer_files,
encoder_path= encoder_files,
vae_path= vae_files,
offload_mode="cpu_offload",
device = device,
)
pipe = QwenImagePipeline.from_pretrained(
config,
)
lora_file = 'Qwen-Image-2512-Turbo-LoRA/Wuli-Qwen-Image-2512-Turbo-LoRA-4steps-V1.0-bf16.safetensors'
pipe.load_lora(
path=lora_file,
scale=1.0,
fused=True,
)
........

Error

File "/Users/abc/models/main.py", line 35, in
pipe.load_lora(
File "/opt/anaconda3/lib/python3.12/site-packages/diffsynth_engine/pipelines/base.py", line 79, in load_lora
self.load_loras([(path, scale)], fused, save_original_weight)
File "/opt/anaconda3/lib/python3.12/site-packages/diffsynth_engine/pipelines/qwen_image.py", line 269, in load_loras
super().load_loras(lora_list, fused, save_original_weight)
File "/opt/anaconda3/lib/python3.12/site-packages/diffsynth_engine/pipelines/base.py", line 76, in load_loras
model.load_loras(lora_args, fused=fused)
File "/opt/anaconda3/lib/python3.12/site-packages/diffsynth_engine/models/base.py", line 44, in load_loras
module = self.get_submodule(key)
^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/anaconda3/lib/python3.12/site-packages/torch/nn/modules/module.py", line 710, in get_submodule
raise AttributeError(
AttributeError: QwenImageDiT has no attribute diffusion_model

[szlzhm]

我也遇到相似问题了，环境：
win10+cuda 128 +torch2.8+python3.10
Qwen-Image-2512 + wuli-art的Qwen-Image-2512-turbo-lora模型

[akaitsuki-ii]

@donaldxu @szlzhm the issue is caused by out-of-date code, please use the new release v0.7.0 instead.