[Bug]: bs_roformer_instrumental_resurrection_unwa.ckpt produces no output files in v0.44.5

[jin-x]

Describe the bug

Version:
audio-separator 0.44.5

Model:
bs_roformer_instrumental_resurrection_unwa.ckpt

Problem:
Running this model directly completes without any exceptions but produces no output files.

Command:

audio-separator --model_filename bs_roformer_instrumental_resurrection_unwa.ckpt input.mp3

There is never a Saving Instrumental stem to ... message, and no output file is created.

This model is also included in the instrumental_clean ensemble preset. Since it produces no output files, the preset reports:

Ensembling 1 stems for type: Instrumental

instead of:

Ensembling 2 stems for type: Instrumental

Additional observations:

• instrumental_full works correctly (2 intermediate files are created and the log reports Ensembling 2 stems).
• karaoke works correctly (3 intermediate files are created and the log reports Ensembling 3 stems).
• Process Monitor confirms that bs_roformer_instrumental_resurrection_unwa.ckpt never creates any intermediate output file.
• The model loads successfully and runs to completion without any exceptions.

Could this be a regression in MDXCSeparator, or is there an issue with this model's YAML/configuration?

Have you searched for existing issues? 🔎

• I have searched and found no existing issues.

Screenshots or Videos

No response

Logs

**Log from the `bs_roformer_instrumental_resurrection_unwa.ckpt` model**

2026-07-25 14:02:56.584 - INFO - cli - Separator version 0.44.5 beginning with input path(s): input.mp3
2026-07-25 14:02:56.585 - INFO - separator - Separator version 0.44.5 instantiating with output_dir: None, output_format: MP3
2026-07-25 14:02:56.585 - INFO - separator - Output directory not specified. Using current working directory.
2026-07-25 14:02:56.585 - INFO - separator - Using model directory from model_file_dir parameter: D:\ML\UVR\models\MDX_Net_Models
2026-07-25 14:02:56.585 - INFO - separator - Operating System: Windows 10.0.26200
2026-07-25 14:02:56.585 - INFO - separator - System: Windows Node: HOME Release: 11 Machine: AMD64 Proc: Intel64 Family 6 Model 151 Stepping 2, GenuineIntel
2026-07-25 14:02:56.586 - INFO - separator - Python Version: 3.13.0
2026-07-25 14:02:56.586 - INFO - separator - PyTorch Version: 2.7.1+cu128
2026-07-25 14:02:56.615 - INFO - separator - FFmpeg installed: ffmpeg version 8.1.1-full_build-www.gyan.dev Copyright (c) 2000-2026 the FFmpeg developers
2026-07-25 14:02:56.616 - INFO - separator - ONNX Runtime GPU package installed with version: 1.27.0
2026-07-25 14:02:56.629 - INFO - separator - CUDA is available in Torch, setting Torch device to CUDA
2026-07-25 14:02:56.629 - INFO - separator - ONNXruntime has CUDAExecutionProvider available, enabling acceleration
2026-07-25 14:02:56.629 - INFO - separator - Loading model bs_roformer_instrumental_resurrection_unwa.ckpt...
2026-07-25 14:02:58.325 - INFO - mdxc_separator - MDXC Separator initialisation complete
2026-07-25 14:02:58.326 - INFO - separator - Roformer loading stats: {'new_implementation_success': 1, 'total_failures': 0}
2026-07-25 14:02:58.326 - INFO - separator - Load model duration: 00:00:01
2026-07-25 14:02:58.326 - INFO - separator - Processing file: input.mp3
2026-07-25 14:02:58.326 - INFO - separator - Starting separation process for audio_file_path: input.mp3
2026-07-25 14:02:58.328 - INFO - common_separator - Input audio subtype: MPEG_LAYER_III
2026-07-25 14:02:58.328 - WARNING - common_separator - Unknown audio subtype MPEG_LAYER_III, defaulting to 16-bit output
2026-07-25 14:02:58.329 - INFO - common_separator - Detected input bit depth: 16-bit
100%|##########| 41/41 [00:21<00:00,  1.88it/s]
2026-07-25 14:03:22.285 - INFO - common_separator - Clearing input audio file paths, sources and stems...
2026-07-25 14:03:22.348 - INFO - separator - Separation duration: 00:00:24
2026-07-25 14:03:22.348 - ERROR - cli - Separation produced no output files — see errors above.


**Log from the `instrumental_clean` ensemble preset:**

2026-07-25 14:06:10.885 - INFO - cli - Separator version 0.44.5 beginning with input path(s): input.mp3
2026-07-25 14:06:10.886 - INFO - separator - Separator version 0.44.5 instantiating with output_dir: None, output_format: MP3
2026-07-25 14:06:10.886 - INFO - separator - Output directory not specified. Using current working directory.
2026-07-25 14:06:10.886 - INFO - separator - Using model directory from model_file_dir parameter: D:\ML\UVR\models\MDX_Net_Models
2026-07-25 14:06:10.887 - INFO - separator - Loaded ensemble preset 'instrumental_clean': Instrumental Clean — Cleanest instrumentals with minimal vocal bleed — Fv7z (bleedless 44.61) + Resurrection Inst (SDR 17.25)
2026-07-25 14:06:10.887 - INFO - separator - Operating System: Windows 10.0.26200
2026-07-25 14:06:10.887 - INFO - separator - System: Windows Node: HOME Release: 11 Machine: AMD64 Proc: Intel64 Family 6 Model 151 Stepping 2, GenuineIntel
2026-07-25 14:06:10.887 - INFO - separator - Python Version: 3.13.0
2026-07-25 14:06:10.887 - INFO - separator - PyTorch Version: 2.7.1+cu128
2026-07-25 14:06:10.916 - INFO - separator - FFmpeg installed: ffmpeg version 8.1.1-full_build-www.gyan.dev Copyright (c) 2000-2026 the FFmpeg developers
2026-07-25 14:06:10.917 - INFO - separator - ONNX Runtime GPU package installed with version: 1.27.0
2026-07-25 14:06:10.932 - INFO - separator - CUDA is available in Torch, setting Torch device to CUDA
2026-07-25 14:06:10.932 - INFO - separator - ONNXruntime has CUDAExecutionProvider available, enabling acceleration
2026-07-25 14:06:10.932 - INFO - separator - Multiple models specified for ensembling: ['mel_band_roformer_instrumental_fv7z_gabox.ckpt', 'bs_roformer_instrumental_resurrection_unwa.ckpt']
2026-07-25 14:06:10.932 - INFO - separator - Ensemble processing for file: input.mp3
2026-07-25 14:06:10.933 - INFO - separator - Processing with model: mel_band_roformer_instrumental_fv7z_gabox.ckpt
2026-07-25 14:06:10.933 - INFO - separator - Loading model mel_band_roformer_instrumental_fv7z_gabox.ckpt...
2026-07-25 14:06:14.674 - INFO - mdxc_separator - MDXC Separator initialisation complete
2026-07-25 14:06:14.674 - INFO - separator - Roformer loading stats: {'new_implementation_success': 1, 'total_failures': 0}
2026-07-25 14:06:14.674 - INFO - separator - Load model duration: 00:00:03
2026-07-25 14:06:14.674 - INFO - separator - Starting separation process for audio_file_path: input.mp3
2026-07-25 14:06:14.677 - INFO - common_separator - Input audio subtype: MPEG_LAYER_III
2026-07-25 14:06:14.677 - WARNING - common_separator - Unknown audio subtype MPEG_LAYER_III, defaulting to 16-bit output
2026-07-25 14:06:14.677 - INFO - common_separator - Detected input bit depth: 16-bit
100%|##########| 41/41 [00:21<00:00,  1.94it/s]
2026-07-25 14:06:53.050 - INFO - common_separator - Clearing input audio file paths, sources and stems...
2026-07-25 14:06:53.100 - INFO - separator - Separation duration: 00:00:21
2026-07-25 14:06:53.100 - INFO - separator - Ensembling 1 stems for type: Instrumental
2026-07-25 14:06:53.418 - INFO - common_separator - Audio duration is 0.09 hours (323.41 seconds).
2026-07-25 14:06:53.418 - INFO - common_separator - Using pydub for writing.
2026-07-25 14:06:53.466 - INFO - common_separator - Writing output with 16-bit depth
2026-07-25 14:06:56.522 - INFO - cli - Separation complete! Output file(s): E:\input_(Instrumental)_preset_instrumental_clean.mp3

System Info

Operating System: Windows 11 Pro 25H2
Python version: 3.13.0

Additional Information

No response

[coderabbitai]

🔗 Related PRs

#256 - Add file-level chunking for large audio files (fixes #158) [closed]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!