black-forest-labs

/

FLUX.1-Fill-dev

like 263

Follow

Black Forest Labs 3,545

So apparently with the new Fill model number of channels changes from the normal 64 to 384. The FluxTransformer2DModel is apparently not geared to accept this. Even with using the parameters below you get an error:

from diffusers import FluxFillPipeline
#base_model = "black-forest-labs/FLUX.1-Fill-dev"
base_model = "./flux-dev/inpaint/flux1-fill-dev.safetensors"
dtype = torch.bfloat16

transformer = FluxTransformer2DModel.from_single_file(base_model, subfolder="transformer", low_cpu_mem_usage=False, ignore_mismatched_sizes=True, torch_dtype=dtype)

Message=Cannot load because x_embedder.weight expected shape tensor(..., device='meta', size=(3072, 64)), but got torch.Size([3072, 384]). If you want to instead overwrite randomly initialized weights, please make sure to pass both low_cpu_mem_usage=False and ignore_mismatched_sizes=True. For more information, see also: https://github.com/huggingface/diffusers/issues/1619#issuecomment-1345604389 as an example.
Source=C:\Users\xxxxx\source\repos\AI\runtimes\bin\windows\Python312\Lib\site-packages\diffusers\models\model_loading_utils.py
StackTrace:
File "C:\Users\xxxxx\source\repos\AI\runtimes\bin\windows\Python312\Lib\site-packages\diffusers\models\model_loading_utils.py", line 223, in load_model_dict_into_meta
raise ValueError(
File "C:\Users\xxxxx\source\repos\AI\runtimes\bin\windows\Python312\Lib\site-packages\diffusers\loaders\single_file_model.py", line 299, in from_single_file
unexpected_keys = load_model_dict_into_meta(model, diffusers_format_checkpoint, dtype=torch_dtype)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\xxxxx\source\repos\AI\runtimes\bin\windows\Python312\Lib\site-packages\huggingface_hub\utils_validators.py", line 114, in _inner_fn
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "C:\Users\xxxxx\source\repos\AI\modules\Inpaint-Anything\app\app.py", line 194, in quantize_model
transformer = FluxTransformer2DModel.from_single_file(base_model, subfolder="transformer", low_cpu_mem_usage=False, ignore_mismatched_sizes=True, torch_dtype=dtype)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\xxxxx\source\repos\AI\modules\Inpaint-Anything\app\app.py", line 404, in setup_model (Current frame)
quantize_model()
File "C:\Users\xxxxx\source\repos\AI\runtimes\bin\windows\Python312\Lib\site-packages\gradio\utils.py", line 826, in wrapper
response = f(*args, **kwargs)
File "C:\Users\xxxxx\source\repos\AI\runtimes\bin\windows\Python312\Lib\site-packages\anyio_backends_asyncio.py", line 859, in run
result = context.run(func, *args)
File "C:\Users\xxxxx\source\repos\AI\runtimes\bin\windows\Python312\Lib\threading.py", line 1075, in _bootstrap_inner
self.run()
File "C:\Users\xxxxx\source\repos\AI\runtimes\bin\windows\Python312\Lib\threading.py", line 1032, in _bootstrap
self._bootstrap_inner()
ValueError: Cannot load because x_embedder.weight expected shape tensor(..., device='meta', size=(3072, 64)), but got torch.Size([3072, 384]). If you want to instead overwrite randomly initialized weights, please make sure to pass both low_cpu_mem_usage=False and ignore_mismatched_sizes=True. For more information, see also: https://github.com/huggingface/diffusers/issues/1619#issuecomment-1345604389 as an example.