[Bug]: Checkpoint merger error

[gee666]

Is there an existing issue for this?

• I have searched the existing issues and checked the recent builds/commits

What happened?

Error when trying to merge checkpoint

Loading C:\Automatic1111\models\Stable-diffusion\realisticVisionV20_v20_full.safetensors...
Loading C:\Automatic1111\models\Stable-diffusion\geetesha\oira666-02-v1.5_7000.safetensors...
Merging...
100%|█████████████████████████████████████████████████████████████████████████████| 1131/1131 [00:02<00:00, 449.38it/s]
Saving to C:\Automatic1111\models\Stable-diffusion\oira666_7000_realisticview.safetensors...
Error loading/saving model file:
Traceback (most recent call last):
  File "C:\Automatic1111\modules\ui.py", line 1722, in modelmerger
    results = modules.extras.run_modelmerger(*args)
  File "C:\Automatic1111\modules\extras.py", line 285, in run_modelmerger
    safetensors.torch.save_file(theta_0, output_modelname, metadata=metadata)
  File "C:\Automatic1111\venv\lib\site-packages\safetensors\torch.py", line 232, in save_file
    serialize_file(_flatten(tensors), filename, metadata=metadata)
TypeError: argument 'metadata': 'dict' object cannot be converted to 'PyString'

Steps to reproduce the problem

1. Go to Checkpoint merger tab
2. Set up two models (I tried a lot of different combinations with the same result)
3. Set up params: m 0.15, Weighted sum, safetensors, A, B or C, Bake in vae: none;
4. Press Merge
5. Wait for a few seconds
6. Receive the error

What should have happened?

Should have merged with no errors

Commit where the problem happens

89f9faa

What platforms do you use to access the UI ?

Windows

What browsers do you use to access the UI ?

Google Chrome

Command Line Arguments

--xformers --api

List of extensions

Auto-Photoshop-StableDiffusion-Plugin
deforum-for-automatic1111-webui
sd-webui-controlnet
sd_dreambooth_extension
LDSR
Lora
ScuNET
SwinIR
prompt-bracket-checker

Console logs

Loading C:\Automatic1111\models\Stable-diffusion\realisticVisionV20_v20_full.safetensors...
Loading C:\Automatic1111\models\Stable-diffusion\geetesha\oira666-02-v1.5_7000.safetensors...
Merging...
100%|█████████████████████████████████████████████████████████████████████████████| 1131/1131 [00:02<00:00, 449.38it/s]
Saving to C:\Automatic1111\models\Stable-diffusion\oira666_7000_realisticview.safetensors...
Error loading/saving model file:
Traceback (most recent call last):
  File "C:\Automatic1111\modules\ui.py", line 1722, in modelmerger
    results = modules.extras.run_modelmerger(*args)
  File "C:\Automatic1111\modules\extras.py", line 285, in run_modelmerger
    safetensors.torch.save_file(theta_0, output_modelname, metadata=metadata)
  File "C:\Automatic1111\venv\lib\site-packages\safetensors\torch.py", line 232, in save_file
    serialize_file(_flatten(tensors), filename, metadata=metadata)
TypeError: argument 'metadata': 'dict' object cannot be converted to 'PyString'

Additional information

No response

[joebbowers]

This has not been resolved as of 10/13/23

`*** Error loading/saving model file
Traceback (most recent call last):
File "C:\stablediffusion\stable-diffusion-webui\modules\ui_checkpoint_merger.py", line 21, in modelmerger
results = extras.run_modelmerger(*args)
File "C:\stablediffusion\stable-diffusion-webui\modules\extras.py", line 315, in run_modelmerger
safetensors.torch.save_file(theta_0, output_modelname, metadata=metadata if len(metadata)>0 else None)
File "C:\stablediffusion\stable-diffusion-webui\venv\lib\site-packages\safetensors\torch.py", line 232, in save_file
serialize_file(_flatten(tensors), filename, metadata=metadata)
TypeError: argument 'metadata': 'dict' object cannot be converted to 'PyString'

---`

[dennisvink]

Same issue

[Webtransformer]

Same here

[mikerodrigues]

Also seeing this issue.

[AvaCadava]

still a problem

[blackorbit1]

Same here

[cvar66]

Just ran into this issue today

[J-Cott]

For me saving as a .ckpt works but not as .safetensors for some models. Annoying to have to convert them in a 3rd party app afterwards.

[joebbowers]

I haven't tried that but I will. I'll be happy enough if that works.

…

On Thu, Dec 7, 2023, 1:30 AM J-Cott ***@***.***> wrote: For me saving as a .ckpt works but not as .safetensors for some models. Annoying to have to convert them in a 3rd party app afterwards. — Reply to this email directly, view it on GitHub <#10460 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A4FOFN62HIE737SH24DNYTDYIGSG7AVCNFSM6AAAAAAYEVF6OKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNBVGE3TOMZWGQ> . You are receiving this because you commented.Message ID: ***@***.***>

[Flo0]

Unchecking "Copy metadata from merged models" helped me avoid this issue.

[Paul-Williams]

Snap :(

For me it only seems to happen for SDXL models

[Paul-Williams]

@Flo0 - Any idea how/if excluding this effects the model?

[armentix]

To exclude this issue, i needed to load each models i wanted to merge first (having the hash tag wrote in the title) then i could merge them afterwards. Don't know why this is needed.

[YashamonSensei]

@Flo0 - Any idea how/if excluding this effects the model?

Shouldn't affect model itself, only information that is passed with it (i.e. authors, how they were made, ...).

As for topic, it seems like it affects only some checkpoints, and since removing metadata fixes the issue, my guess is that some models have metadata Checkpoint Merger can't parse.