`device_map="auto"` supported for diffusers pipelines?

[johannaSommer]

Describe the bug

Hey dear diffusers team,

for DiffusionPipline, as I understand (hopefully correctly) from this part of the documentation, it should be possible to specify device_map="auto" when loading a pipeline with from_pretrained but this results in a value error saying that this is not supported.

However, the documentation on device placement currently states that only the "balanced" strategy is supported.

Is this possibly similar to #11432 and should be removed from the docstrings / documentation? Happy to help on this with a PR if it turns out to be a mistake in the documentation.

Thanks a lot for your hard work!

Reproduction

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

or

from diffusers import StableDiffusionPipeline
pipe = StableDiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", device_map="auto")

Logs

---------------------------------------------------------------------------
NotImplementedError                       Traceback (most recent call last)
Cell In[12], line 3
      1 from diffusers import StableDiffusionPipeline
----> 3 pipe = StableDiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", device_map="auto")

File ~/miniconda3/envs/pruna/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py:114, in validate_hf_hub_args.<locals>._inner_fn(*args, **kwargs)
    111 if check_use_auth_token:
    112     kwargs = smoothly_deprecate_use_auth_token(fn_name=fn.__name__, has_token=has_token, kwargs=kwargs)
--> 114 return fn(*args, **kwargs)

File ~/miniconda3/envs/pruna/lib/python3.10/site-packages/diffusers/pipelines/pipeline_utils.py:745, in DiffusionPipeline.from_pretrained(cls, pretrained_model_name_or_path, **kwargs)
    742     raise ValueError("`device_map` must be a string.")
    744 if device_map is not None and device_map not in SUPPORTED_DEVICE_MAP:
--> 745     raise NotImplementedError(
    746         f"{device_map} not supported. Supported strategies are: {', '.join(SUPPORTED_DEVICE_MAP)}"
    747     )
    749 if device_map is not None and device_map in SUPPORTED_DEVICE_MAP:
    750     if is_accelerate_version("<", "0.28.0"):

NotImplementedError: auto not supported. Supported strategies are: balanced

System Info

• 🤗 Diffusers version: 0.33.1
• Platform: Linux-5.15.0-139-generic-x86_64-with-glibc2.35
• Running on Google Colab?: No
• Python version: 3.10.16
• PyTorch version (GPU?): 2.7.0+cu126 (True)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Huggingface_hub version: 0.30.2
• Transformers version: 4.51.3
• Accelerate version: 1.6.0
• PEFT version: 0.15.2
• Bitsandbytes version: 0.45.5
• Safetensors version: 0.5.3
• xFormers version: not installed
• Accelerator: NVIDIA H100 PCIe, 81559 MiB
  NVIDIA H100 PCIe, 81559 MiB
• Using GPU in script?: yes
• Using distributed or parallel set-up in script?: yes

Who can help?

No response

[DN6]

Hi @johannaSommer yes, great catch again 👍🏽. It is indeed a mistake in the docs. Would it be possible for you to open a PR to fix?

[johannaSommer]

@DN6 thanks for having a look! I will open a PR yes 🙌

[DN6]

Or actually @johannaSommer , let me first look into whether we can just quickly add support for "auto"

[johannaSommer]

Sure, sounds good! Again, happy to help wherever you think it makes sense.