Sana-sprint pipeline failing on MPS

[bertaveira]

Describe the bug

I noticed that if I run the SanaSprintPipeline on MPS the output is totally broken. Running on CPU works fine.

Reproduction

from diffusers.pipelines.sana.pipeline_sana_sprint import SanaSprintPipeline

pipeline = SanaSprintPipeline.from_pretrained("Efficient-Large-Model/Sana_Sprint_0.6B_1024px_diffusers")
pipeline.to("mps")

prompt = "a tiny astronaut hatching from an egg on the moon"
image = pipeline(prompt=prompt).images[0]
image.show()

The output I get is this:

Just by removing the pipeline.to("mps") it generates a good image:

Logs

System Info

Macbook M2 Pro 32GB RAM

• 🤗 Diffusers version: 0.33.0.dev0
• Platform: macOS-14.7.4-arm64-arm-64bit
• Running on Google Colab?: No
• Python version: 3.12.8
• PyTorch version (GPU?): 2.6.0 (False)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Huggingface_hub version: 0.29.3
• Transformers version: 4.49.0
• Accelerate version: 1.5.2
• PEFT version: 0.15.0
• Bitsandbytes version: not installed
• Safetensors version: 0.5.3
• xFormers version: not installed
• Accelerator: Apple M2 Pro
• Using GPU in script?: yes?
• Using distributed or parallel set-up in script?: No

Who can help?

@yiyixuxu @DN6

[Vargol]

I ran your script on my M3 iMac and it generated an image fine.
There's a very good chance that it's an issue with torch running on your 'old' version of MacOS, they tend to test things on the latest MacOS version (and have occasionally been told issues I've raised are fixed in a Beta MacOS).

script ran...

from diffusers.pipelines.sana.pipeline_sana_sprint import SanaSprintPipeline

pipeline = SanaSprintPipeline.from_pretrained("Efficient-Large-Model/Sana_Sprint_0.6B_1024px_diffusers")
pipeline.to("mps")

prompt = "a tiny astronaut hatching from an egg on the moon"
image = pipeline(prompt=prompt).images[0]
image.show()

output...

[bertaveira]

I can confirm this is a MacOS issue. Updated from Sonoma to Sequoia and it now works without nay change or dependency update at all, and it works now.