Docker-Compose-File not working

[Husky110]

Hi - The current state (just pulled the repository) seems to have an error in the docker-compose.yaml file.
I think this refers to the PR #171 and issue #150.
I cloned the repo, ran docker-compose build, ran docker-compose up and hit "Generate" to download the model and then this happened:

ace-step  | 
ace-step  | ==========
ace-step  | == CUDA ==
ace-step  | ==========
ace-step  | 
ace-step  | CUDA Version 12.6.0
ace-step  | 
ace-step  | Container image Copyright (c) 2016-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
ace-step  | 
ace-step  | This container image and its contents are governed by the NVIDIA Deep Learning Container License.
ace-step  | By pulling and using the container, you accept the terms and conditions of this license:
ace-step  | https://developer.nvidia.com/ngc/nvidia-deep-learning-container-license
ace-step  | 
ace-step  | A copy of this license is made available in this container at /NGC-DL-CONTAINER-LICENSE for your convenience.
ace-step  | 
ace-step  | * Running on local URL:  http://0.0.0.0:7865
ace-step  | * To create a public link, set `share=True` in `launch()`.
ace-step  | 2025-05-25 11:16:08.029 | WARNING  | acestep.pipeline_ace_step:__call__:1481 - Checkpoint not loaded, loading checkpoint...
ace-step  | 2025-05-25 11:16:08.029 | INFO     | acestep.pipeline_ace_step:get_checkpoint_path:180 - Download models from Hugging Face: ACE-Step/ACE-Step-v1-3.5B, cache to: /home/appuser/.cache/ace-step/checkpoints
ace-step  | /opt/venv/lib/python3.10/site-packages/torch/nn/utils/weight_norm.py:143: FutureWarning: `torch.nn.utils.weight_norm` is deprecated in favor of `torch.nn.utils.parametrizations.weight_norm`.
ace-step  |   WeightNorm.apply(module, name, dim)
ace-step  | 2025-05-25 11:37:16.151 | INFO     | acestep.pipeline_ace_step:__call__:1489 - Model loaded in 1268.12 seconds.
ace-step  | 2025-05-25 11:37:16.300 | INFO     | acestep.pipeline_ace_step:text2music_diffusion_process:851 - cfg_type: apg, guidance_scale: 15, omega_scale: 10
ace-step  | 2025-05-25 11:37:16.303 | INFO     | acestep.pipeline_ace_step:text2music_diffusion_process:1076 - start_idx: 15, end_idx: 45, num_inference_steps: 60
100%|██████████| 60/60 [00:58<00:00,  1.03it/s]
  0%|          | 0/1 [00:00<?, ?it/s]2025-05-25 11:38:20.591 | WARNING  | acestep.pipeline_ace_step:save_wav_file:1385 - save_path is None, using default path ./outputs/
ace-step  | 2025-05-25 11:38:20.591 | INFO     | acestep.pipeline_ace_step:save_wav_file:1400 - Saving audio to ./outputs/output_20250525113820_0.wav
  0%|          | 0/1 [00:00<?, ?it/s]
ace-step  | Traceback (most recent call last):
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/gradio/queueing.py", line 625, in process_events
ace-step  |     response = await route_utils.call_process_api(
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/gradio/route_utils.py", line 322, in call_process_api
ace-step  |     output = await app.get_blocks().process_api(
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/gradio/blocks.py", line 2191, in process_api
ace-step  |     result = await self.call_function(
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/gradio/blocks.py", line 1702, in call_function
ace-step  |     prediction = await anyio.to_thread.run_sync(  # type: ignore
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/anyio/to_thread.py", line 56, in run_sync
ace-step  |     return await get_async_backend().run_sync_in_worker_thread(
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/anyio/_backends/_asyncio.py", line 2470, in run_sync_in_worker_thread
ace-step  |     return await future
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/anyio/_backends/_asyncio.py", line 967, in run
ace-step  |     result = context.run(func, *args)
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/gradio/utils.py", line 894, in wrapper
ace-step  |     response = f(*args, **kwargs)
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/acestep/pipeline_ace_step.py", line 1665, in __call__
ace-step  |     output_paths = self.latents2audio(
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/acestep/cpu_offload.py", line 32, in wrapper
ace-step  |     return func(self, *args, **kwargs)
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/acestep/pipeline_ace_step.py", line 1371, in latents2audio
ace-step  |     output_audio_path = self.save_wav_file(
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/acestep/pipeline_ace_step.py", line 1401, in save_wav_file
ace-step  |     torchaudio.save(
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/torchaudio/_backend/utils.py", line 313, in save
ace-step  |     return backend.save(
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/torchaudio/_backend/soundfile.py", line 44, in save
ace-step  |     soundfile_backend.save(
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/torchaudio/_backend/soundfile_backend.py", line 457, in save
ace-step  |     soundfile.write(file=filepath, data=src, samplerate=sample_rate, subtype=subtype, format=format)
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/soundfile.py", line 363, in write
ace-step  |     with SoundFile(file, 'w', samplerate, channels,
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/soundfile.py", line 690, in __init__
ace-step  |     self._file = self._open(file, mode_int, closefd)
ace-step  |   File "/opt/venv/lib/python3.10/site-packages/soundfile.py", line 1265, in _open
ace-step  |     raise LibsndfileError(err, prefix="Error opening {0!r}: ".format(self.name))
ace-step  | soundfile.LibsndfileError: Error opening './outputs/output_20250525113820_0.wav': System error.

I think those relative paths might not be the best idea for docker tho and the - ACE_OUTPUT_DIR=/app/outputs-environment seems to be not respected. Please investigate.
System: Arch Linux

[Husky110]

Additional: I found the problem in the pipeline_ace_step.py-file on line 1384. Seems like the savepath does not recognize the docker-changes.

[darkcl]

i encounter the same problem of this docker-compose file, there are multiple problems:

• Dockerfile create a new user 1001, it causes the System error in your log (the user id that want to write file does not have the permission to do so), i don't know why they do it this way though....
• It misses dependency for ffmpeg and torchcodec
• It won't start because of a weird dependency mis-match of cuda

RuntimeError: Detected that PyTorch and TorchAudio were compiled with different CUDA versions. PyTorch has CUDA version 12.8 whereas TorchAudio has CUDA version 12.6. Please install the TorchAudio version that matches your PyTorch version.

i created a fork as a temp fix and it runs on my machine (i am also using arch)

main...darkcl:ACE-Step:main