Applio icon indicating copy to clipboard operation
Applio copied to clipboard

[Bug]: Conversion fails when index file is selected despite successful run without index

Open nimishdm opened this issue 3 months ago • 9 comments

Project Version

3.4.0

Platform and OS Version

Mac OS, 15.6.1

Affected Devices

MacBook Pro

Existing Issues

No response

What happened?

I downloaded a model from HuggingFace and added both the .pth file and the .index file to the application by drag and drop. Both files appear correctly in the dropdown menus for selecting the model and the index.

When I run a conversion without selecting the index, the conversion completes successfully. However, when I select the index from the dropdown and run the conversion on the same file, it fails.

Environment • OS: macOS 15.6.1

Steps to reproduce

  1. Download the model from HuggingFace - https://huggingface.co/Astral-P/Ryuji/resolve/main/Ryuji.zip.
  2. Add both the .pth and .index files to the application via drag and drop.
  3. Select the .pth file from the model dropdown.
  4. Run a conversion → succeeds.
  5. Select the .index file in addition to the .pth file.
  6. Run a conversion on the same file → fails.

Expected behavior

Conversion should succeed when using the index file, just like it does without the index.

Attachments

No response

Screenshots or Videos

https://github.com/user-attachments/assets/dc6add6b-fa3e-44a2-aedb-ebae514a0b12

Additional Information

No response

nimishdm avatar Sep 14 '25 04:09 nimishdm

Are you sure you’re running the inference with a valid .index file? You didn’t happen to rename the trained one to added, did you?

Vidalnt avatar Sep 14 '25 12:09 Vidalnt

Are you sure you’re running the inference with a valid .index file? You didn’t happen to rename the trained one to added, did you?

I used the downloaded file as is, didn't rename it. It's not specific to this model. It fails whenever I use an index file with any model.

nimishdm avatar Sep 14 '25 12:09 nimishdm

Is there any additional error shown in the console?

Vidalnt avatar Sep 14 '25 12:09 Vidalnt

I just tested the inference on the latest version, and I don’t have any issues.

Vidalnt avatar Sep 14 '25 12:09 Vidalnt

Is there any additional error shown in the console?

It says segmentation fault in the console. No other errors.

I just tested the inference on the latest version, and I don’t have any issues.

Did you try on MacOS? My machine is an M2 Pro MacBook Pro. I cloned the repository only yesterday. I guess I was running the latest version.

nimishdm avatar Sep 14 '25 12:09 nimishdm

macOS..., unfortunately I don’t have one and I don’t think I’ll be able to find a solution for you at the moment. Still, if version 3.4.0 doesn’t work for you, I wouldn’t know if 3.5.0 would make a difference. Also, I’ve never seen the 'segmentation fault' error before, I’ll have to investigate what could possibly cause it, but I can’t promise anything.

Vidalnt avatar Sep 14 '25 12:09 Vidalnt

Still, for the moment, if you’d like, you can try the notebooks until a solution is found.

Vidalnt avatar Sep 14 '25 12:09 Vidalnt

Did you try on MacOS? My machine is an M2 Pro MacBook Pro. I cloned the repository only yesterday. I guess I was running the latest version.

you need to do the usual

export PYTORCH_ENABLE_MPS_FALLBACK=1 export PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0 export OMP_NUM_THREADS=1

AznamirWoW avatar Sep 14 '25 15:09 AznamirWoW

@AznamirWoW Thanks a lot. Adding export OMP_NUM_THREADS=1 to run-applio.sh solved the issue for me. Could you please add this fix to the repo? Also does this fix limit the performance?

@Vidalnt

nimishdm avatar Sep 20 '25 06:09 nimishdm