-
Notifications
You must be signed in to change notification settings - Fork 6.6k
Closed
Labels
bugSomething isn't workingSomething isn't working
Description
Describe the bug
num_in_channels has no effect when call StableDiffusionXLInpaintPipeline.from_single_file and StableDiffusionXLPipeline.from_single_file
Reproduction
- download normal sdxl model from here: https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/sd_xl_base_1.0.safetensors
- download inpainting sdxl model from here: https://huggingface.co/wangqyqq/sd_xl_base_1.0_inpainting_0.1.safetensors/blob/main/sd_xl_base_1.0_inpainting_0.1.safetensors
Running following code can produce the issue:
from diffusers import StableDiffusionXLInpaintPipeline, StableDiffusionXLPipeline
# load normal sdxl model
pipe = StableDiffusionXLPipeline.from_single_file("sd_xl_base_1.0.safetensors", local_files_only=False, load_safety_checker=False)
print("load normal sdxl model, unet in_channels: ", pipe.unet.config.in_channels)
# load normal sdxl model, by setting num_in_channels=9, it should raise error
pipe = StableDiffusionXLPipeline.from_single_file("sd_xl_base_1.0.safetensors", local_files_only=False, load_safety_checker=False, num_in_channels=9)
print("load normal sdxl model with num_in_channels=9, should raise error, but unet in_channels is: ", pipe.unet.config.in_channels)
print("""This should load ok, but will raise error: ValueError: Trying to set a tensor of shape torch.Size([320, 9, 3, 3]) in "weight" (which has shape torch.Size([320, 4, 3, 3])), this look incorrect.""")
pipe = StableDiffusionXLInpaintPipeline.from_single_file("sd_xl_base_1.0_inpainting_0.1.safetensors", local_files_only=False, load_safety_checker=False, num_in_channels=9)The original configuration file used by SDXL is using network_config instead of unet_config, which suggests that there may be an issue with this line of code:
| if "unet_config" in original_config["model"]["params"]: |
Logs
No response
System Info
diffusersversion: 0.25.0- Platform: macOS-13.5-arm64-arm-64bit
- Python version: 3.8.10
- PyTorch version (GPU?): 2.1.0 (False)
- Huggingface_hub version: 0.20.2
- Transformers version: 4.36.2
- Accelerate version: 0.24.1
- xFormers version: not installed
- Using GPU in script?:
- Using distributed or parallel set-up in script?: <fill in
Who can help?
Metadata
Metadata
Assignees
Labels
bugSomething isn't workingSomething isn't working