AttributeError: 'Parameter' object has no attribute 'named_children'

[klyuhang9]

Describe the bug

Hi, friends, I meet a problem I hope to get your help.

When I run the code as follow:

`from diffusers import StableDiffusionPipeline
import torch

pipe = StableDiffusionPipeline.from_pretrained("PATH_OF_MODEL/stable-diffusion",device_map="auto",torch_type=torch.float16, revision="fp16")
`
I save the model in the path "PATH_OF_MODEL/stable-diffusion" from the https://huggingface.co/CompVis/stable-diffusion-v1-4/tree/main
Then, I get the error as follow:

ftfy or spacy is not installed using BERT BasicTokenizer instead of ftfy. Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/home/klyuhang9/miniconda3/envs/PY_TEST/lib/python3.9/site-packages/diffusers/pipeline_utils.py", line 517, in from_pretrained loaded_sub_model = load_method(os.path.join(cached_folder, name), **loading_kwargs) File "/home/klyuhang9/miniconda3/envs/PY_TEST/lib/python3.9/site-packages/transformers/modeling_utils.py", line 2265, in from_pretrained device_map = infer_auto_device_map( File "/home/klyuhang9/miniconda3/envs/PY_TEST/lib/python3.9/site-packages/accelerate/utils/modeling.py", line 480, in infer_auto_device_map max_layer_size, max_layer_names = get_max_layer_size(modules_to_treat, module_sizes, no_split_module_classes) File "/home/klyuhang9/miniconda3/envs/PY_TEST/lib/python3.9/site-packages/accelerate/utils/modeling.py", line 261, in get_max_layer_size modules_children = list(module.named_children()) AttributeError: 'Parameter' object has no attribute 'named_children'
My environment: python=3.9.13, diffusers=0.5.1 torch=1.12.1, transformers = 4.12.1, spacy = 3.4.1,accelerate=0.13.1 which run on ubuntu=22.04.

I don't know why, I sincerely to get your team help.Thanks,bro!

Reproduction

`from diffusers import StableDiffusionPipeline
import torch

pipe = StableDiffusionPipeline.from_pretrained("PATH_OF_MODEL/stable-diffusion",device_map="auto",torch_type=torch.float16, revision="fp16")'

Logs

`ftfy or spacy is not installed using BERT BasicTokenizer instead of ftfy.
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/klyuhang9/miniconda3/envs/PY_TEST/lib/python3.9/site-packages/diffusers/pipeline_utils.py", line 517, in from_pretrained
    loaded_sub_model = load_method(os.path.join(cached_folder, name), **loading_kwargs)
  File "/home/klyuhang9/miniconda3/envs/PY_TEST/lib/python3.9/site-packages/transformers/modeling_utils.py", line 2265, in from_pretrained
    device_map = infer_auto_device_map(
  File "/home/klyuhang9/miniconda3/envs/PY_TEST/lib/python3.9/site-packages/accelerate/utils/modeling.py", line 480, in infer_auto_device_map
    max_layer_size, max_layer_names = get_max_layer_size(modules_to_treat, module_sizes, no_split_module_classes)
  File "/home/klyuhang9/miniconda3/envs/PY_TEST/lib/python3.9/site-packages/accelerate/utils/modeling.py", line 261, in get_max_layer_size
    modules_children = list(module.named_children())
AttributeError: 'Parameter' object has no attribute 'named_children'

System Info

My environment: python=3.9.13, diffusers=0.5.1 torch=1.12.1, transformers = 4.12.1, spacy = 3.4.1,accelerate=0.13.1 which run on ubuntu=22.04.

[patrickvonplaten]

Hey @klyuhang9,

Thanks a lot for the issue. This is a bug that we recently fixed in accelerate here: huggingface/accelerate#747

I'm not exactly sure when a new accelerate version will come out - also cc @sgugger @muellerzr here (not sure if it's worth making a patch release for this)
Until then could you try the same command with the "main" version of accelerate:

pip install git+https://github.com/huggingface/accelerate.git

?

[sgugger]

I can put it in a patch later today!

[sgugger]

Patch is out on Pypi.

[patrickvonplaten]

Thanks a lot! @klyuhang9 doing:

pip install --upgrade accelerate
pip install --upgrade transformers
pip install --upgrade diffusers

should make your code above run correctly

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.