TypeError: CLIPTextModel.__init__() got an unexpected keyword argument 'offload_state_dict'

[zillur01]

Describe the bug

Getting this TypeError: CLIPTextModel.__init__() got an unexpected keyword argument 'offload_state_dict' when running a DiffusionPipeline.from_pretrained

Reproduction

I was following this https://huggingface.co/docs/diffusers/main/en/installation for installation. This is what I have done so far:

conda create -n diffusion python=3.10
conda activate diffusion
pip install diffusers["torch"] transformers

Then I tried to start a pipeline by running:

from diffusers import DiffusionPipeline
pipeline = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", use_safetensors=True)

Logs

>>> from diffusers import DiffusionPipeline
>>> pipeline = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", use_safetensors=True)
model_index.json: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████| 541/541 [00:00<00:00, 4.34MB/s]
scheduler_config.json: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████| 308/308 [00:00<00:00, 3.30MB/s]
config.json: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 617/617 [00:00<00:00, 166kB/s]
preprocessor_config.json: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████| 342/342 [00:00<00:00, 336kB/s]
special_tokens_map.json: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████| 472/472 [00:00<00:00, 378kB/s]
config.json: 4.72kB [00:00, 10.2MB/s]                                                                                                   | 1/15 [00:00<00:08,  1.65it/s]
merges.txt: 525kB [00:00, 5.92MB/s]
tokenizer_config.json: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████| 806/806 [00:00<00:00, 5.94MB/s]
config.json: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 743/743 [00:00<00:00, 933kB/s]
config.json: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 547/547 [00:00<00:00, 310kB/s]
vocab.json: 1.06MB [00:00, 10.7MB/s]                                                                                                         | 0.00/547 [00:00<?, ?B/s]
safety_checker/model.safetensors: 100%|████████████████████████████████████████████████████████████████████████████████████████████| 1.22G/1.22G [00:05<00:00, 233MB/s]
unet/diffusion_pytorch_model.safetensors: 100%|████████████████████████████████████████████████████████████████████████████████████| 3.44G/3.44G [00:06<00:00, 566MB/s]
vae/diffusion_pytorch_model.safetensors: 100%|██████████████████████████████████████████████████████████████████████████████████████| 335M/335M [00:06<00:00, 55.1MB/s]
text_encoder/model.safetensors: 100%|███████████████████████████████████████████████████████████████████████████████████████████████| 492M/492M [00:06<00:00, 73.1MB/s]
Fetching 15 files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████| 15/15 [00:07<00:00,  2.03it/s]
Loading pipeline components...:  14%|██████████████▎                                                                                     | 1/7 [00:00<00:00, 18.78it/s]
Traceback (most recent call last):0%|████████████████████████████████████████████████████████████████████████████████████████████████| 492M/492M [00:06<00:00, 212MB/s]
  File "<stdin>", line 1, in <module>
  File "/home/zillur/miniconda3/envs/diffusion/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py", line 114, in _inner_fn
    return fn(*args, **kwargs)
  File "/home/zillur/miniconda3/envs/diffusion/lib/python3.10/site-packages/diffusers/pipelines/pipeline_utils.py", line 1025, in from_pretrained
    loaded_sub_model = load_sub_model(
  File "/home/zillur/miniconda3/envs/diffusion/lib/python3.10/site-packages/diffusers/pipelines/pipeline_loading_utils.py", line 849, in load_sub_model
    loaded_sub_model = load_method(os.path.join(cached_folder, name), **loading_kwargs)
  File "/home/zillur/miniconda3/envs/diffusion/lib/python3.10/site-packages/transformers/modeling_utils.py", line 277, in _wrapper
    return func(*args, **kwargs)
  File "/home/zillur/miniconda3/envs/diffusion/lib/python3.10/site-packages/transformers/modeling_utils.py", line 4974, in from_pretrained
    model = cls(config, *model_args, **model_kwargs)
TypeError: CLIPTextModel.__init__() got an unexpected keyword argument 'offload_state_dict'

System Info

Python version: 3.10.18
PyTorch version: 2.8.0
CUDA runtime: 12.8.90
huggingface-hub: 0.35.3
OS: Ubuntu 24.04.3 LTS
diffusers version: 0.35.1
transformers version: 4.57.0
accelerate version: 1.10.1

Who can help?

@sayakpaul @DN6

[zillur01]

downgraded transformers to 4.49 and diffusers to 0.32.2 and it's working now

[moha23]

but doesnt the file require dev version of diffusers? for me it fails while running the file.. saying it needs dev version

[sayakpaul]

We will take care of fixing the issue:

TypeError: CLIPTextModel.__init__() got an unexpected keyword argument 'offload_state_dict'

but doesnt the file require dev version of diffusers? for me it fails while running the file.. saying it needs dev version

You can comment out the line that enforces that requirement for now.

[thuthaouelk10]

downgraded transformers to 4.49 and diffusers to 0.32.2 and it's working now

I faced the same error when running a workflow using the node Hunyuan3D-Wrapper, I downgraded as you said and it works perfecty! Thank you so much!

[sayakpaul]

Can you all check #12438?

[moha23]

Can you all check #12438?

works! thanks

[juanmartinez-fortis]

Happening to all the models using CLIP Text, including Qwen Image Edit. And we can't downgrade to 0.32.2 as Qwen Image Edit is not there. Using the branch right now but this is super crafty

[baizhu0414]

I also encountered this problem, but what makes me uncomfortable is that my current script requires the version of diffusers:

# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
check_min_version("0.36.0.dev0")

Does anyone have solusions?

[baizhu0414]

I also encountered this problem, but what makes me uncomfortable is that my current script requires the version of diffusers:

Will error if the minimal version of diffusers is not installed. Remove at your own risks.

check_min_version("0.36.0.dev0")
Does anyone have solusions?

#12436 (comment)
this is the right method.

[asomoza]

I'll revisit that training script, it's really old, both SD1.5 and SDXL are ancient technologies in the AI space, so those scripts are actually just fine by using older versions of all the libraries. We probably should pin the libraries to the latest working versions and remove the using from main line and comment in the instructions.

Just to clarify:

# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
check_min_version("0.36.0.dev0")

It is fine if you remove it and install older versions of diffusers, transformers and peft for scripts that haven’t been updated in a long time.

[sayakpaul]

Please use #12438, for now.

git clone https://github.com/huggingface/diffusers/
cd diffusers
pip uninstall -y diffusers
pip install -e .

[sayakpaul]

#12438 should fix it.