Cant use my LoRA on colab, getting Error while deserializing header: MetadataIncompleteBuffer (tested with multiple loras)

[nkls2000]

Hey! I´ve been training my own model, and I have trained them with both Autotrain (https://colab.research.google.com/github/huggingface/autotrain-advanced/blob/main/colabs/AutoTrain.ipynb)

and with this one:
https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/SDXL_DreamBooth_LoRA_.ipynb

The training went fine, and my model works when I test them on Hugging face.
I now want to run my own models in Foocus, and I´m trying in the colab version here:
https://colab.research.google.com/github/lllyasviel/Fooocus/blob/main/fooocus_colab.ipynb

I´m getting an error, and its the same error with all my models. It says
ValueError: Error while deserializing header: MetadataIncompleteBuffer
and then it says my model is corrupt. Would be really happy if anyone could help me with this. How do I solve this?

[nkls2000]

Here is the full log:

`[Parameters] Adaptive CFG = 7
[Parameters] Sharpness = 2
[Parameters] ControlNet Softness = 0.25
[Parameters] ADM Scale = 1.5 : 0.8 : 0.3
[Parameters] CFG = 4.0
[Parameters] Seed = 3308499627304362711
[Parameters] Sampler = dpmpp_2m_sde_gpu - karras
[Parameters] Steps = 30 - 15
[Fooocus] Initializing ...
[Fooocus] Loading models ...
Refiner unloaded.
Request to load LoRAs [['aida-robot.safetensors', 0.99], ['None', 1.0], ['None', 1.0], ['None', 1.0], ['None', 1.0]] for model [/content/Fooocus/models/checkpoints/juggernautXL_v8Rundiffusion.safetensors].
Traceback (most recent call last):
File "/content/Fooocus/modules/patch.py", line 465, in loader
result = original_loader(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/safetensors/torch.py", line 259, in load_file
with safe_open(filename, framework="pt", device=device) as f:
safetensors_rust.SafetensorError: Error while deserializing header: MetadataIncompleteBuffer

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/content/Fooocus/modules/async_worker.py", line 913, in worker
handler(task)
File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/content/Fooocus/modules/async_worker.py", line 429, in handler
pipeline.refresh_everything(refiner_model_name=refiner_model_name, base_model_name=base_model_name,
File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/content/Fooocus/modules/default_pipeline.py", line 236, in refresh_everything
refresh_loras(loras, base_model_additional_loras=base_model_additional_loras)
File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/content/Fooocus/modules/default_pipeline.py", line 134, in refresh_loras
model_base.refresh_loras(loras + base_model_additional_loras)
File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/content/Fooocus/modules/core.py", line 95, in refresh_loras
lora_unmatch = ldm_patched.modules.utils.load_torch_file(lora_filename, safe_load=False)
File "/content/Fooocus/ldm_patched/modules/utils.py", line 13, in load_torch_file
sd = safetensors.torch.load_file(ckpt, device=device.type)
File "/content/Fooocus/modules/patch.py", line 481, in loader
raise ValueError(exp)
ValueError: Error while deserializing header: MetadataIncompleteBuffer
File corrupted: /content/Fooocus/models/loras/aida-robot.safetensors
Fooocus has tried to move the corrupted file to /content/Fooocus/models/loras/aida-robot.safetensors.corrupted
You may try again now and Fooocus will download models again.

Total time: 0.05 seconds`

[nkls2000]

here is a url to our model as well:
https://www.dropbox.com/scl/fi/mkuwpedsxbrapt8128uf5/aida-robot.safetensors?rlkey=4otq83i4z1hd72rwuvgrvzg19&raw=1

Model description
LoRA adaption weights for stabilityai/stable-diffusion-xl-base-1.0.

The weights were trained using DreamBooth.

LoRA for the text encoder was enabled: False.

Special VAE used for training: madebyollin/sdxl-vae-fp16-fix.

Trigger words
You should use a photo of TOK Aida-robot to trigger the image generation.

[mashb1t]

@nkls2000 Relates to AUTOMATIC1111/stable-diffusion-webui#10199.
Which algorithm / settigns have you used to train the LoRA/LyCORIS/LoCon and is there metadata included?

[nkls2000]

I used this colab, with the default settings:
https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/SDXL_DreamBooth_LoRA_.ipynb?authuser=1
(as mentioned above I tried the autotrain one too with default settings, same problem)

[mashb1t]

I've personally neither encountered this issue nor used that playbook, it's possible that it does not produce correct LoRAs or is misconfigured somehow. Maybe you can try to download the file again, it's not impossible that the file has been corrupted on download.
Maybe somebody else had that problem before, please consult Google / Reddit / StackOverflow or other AI forums for potential solutions.

[nkls2000]

I´ve tried to download it multiple times, but the error is persistant. Do you have any suggestion for another colab where I can train my model and see if that works better?

[mashb1t]

Not Colab, but i've used kohya_ss to train various LoRAs. There might be working Colab playbooks out there though, also didn't have a deeper look into the issue.

[nkls2000]

we just tried with this one as well:
https://colab.research.google.com/github/hollowstrawberry/kohya-colab/blob/main/Lora_Trainer_XL.ipynb#scrollTo=OglZzI_ujZq-

and we are still getting the same output :/