pipx installation fails with python 3.13, works with 3.12

[hmijail]

Describe the bug

pipx install "audio-separator[cpu]" fails; the main reason seems to be

The conflict is caused by:
    audio-separator[cpu] 0.28.5 depends on torch<2.5 and >=2.3
...

However, pipx install "audio-separator[cpu]" --python python3.12 installs OK.

audio-separator --env_info
2025-01-05 16:16:01,096 - INFO - separator - Separator version 0.28.5 instantiating with output_dir: None, output_format: WAV
2025-01-05 16:16:01,096 - INFO - separator - Output directory not specified. Using current working directory.
2025-01-05 16:16:01,096 - INFO - separator - Operating System: Darwin Darwin Kernel Version 24.2.0: Fri Dec  6 19:01:59 PST 2024; root:xnu-11215.61.5~2/RELEASE_ARM64_T6000
2025-01-05 16:16:01,107 - INFO - separator - System: Darwin Node: HMMBP.local Release: 24.2.0 Machine: arm64 Proc: arm
2025-01-05 16:16:01,107 - INFO - separator - Python Version: 3.12.7
2025-01-05 16:16:01,108 - INFO - separator - PyTorch Version: 2.4.1
2025-01-05 16:16:01,347 - INFO - separator - FFmpeg installed: ffmpeg version 7.1 Copyright (c) 2000-2024 the FFmpeg developers
2025-01-05 16:16:01,348 - INFO - separator - ONNX Runtime CPU package installed with version: 1.20.1
2025-01-05 16:16:01,380 - INFO - separator - Apple Silicon MPS/CoreML is available in Torch and processor is ARM, setting Torch device to MPS
2025-01-05 16:16:01,380 - INFO - separator - ONNXruntime has CoreMLExecutionProvider available, enabling acceleration

Have you searched for existing issues? 🔎

• I have searched and found no existing issues.

Screenshots or Videos

No response

Logs

No response

System Info

MacOS 15.2 on Apple Silicon M1 Pro
Python 3.13
Pipx 1.7.1

Additional Information

No response

[Bebra777228]

Not all packages of this tool support Python version 3.13, making installation on it impossible.

Supported versions: 3.10 — 3.12

[prasannapattam]

Can numpy and torch be upgraded to the latest versions. Numpy is pretty old and is conflicting with other packages.

numpy = ">=1.23,<2"
torch = ">=2.3,<2.5"

[Margen67]

onnx 1.18 will have Python 3.13 support: onnx/onnx#6339

[beveradb]

As a general rule, if someone wants to try and get audio-separator working with a new Python version, or with relaxed dependency requirements, a PR would certainly be welcomed and merged by me once it's ready!

However, I'm only willing to merge it if we've actually tested that it works, on all currently supported platforms (at the very least amd64 linux and arm64 mac), with all supported model architectures (see audio-separator -l.

PR #188 doesn't (yet) do that - there are failing CI tests, and as far as I could see, there is no description or comment or other evidence that audio-separator actually works (with all supported model architectures) on Python 3.13.

If you're willing to fix the CI tests and do some testing to ensure separation actually works for MDX, MDXC, VR, Roformer etc. with Python 3.13, @Margen67 , please do so and comment your findings in the PR description/comments 🙇

[beveradb]

FYI, audio-separator version 0.31 is now released and should work on all platforms with all supported python versions including 3.13 🙏

I ended up having to fork onnx2torch (https://pypi.org/project/onnx2torch-py313/) to change it to require onnx-weekly instead of onnx for now as otherwise the transitive dependency forced pip to try/fail to build onnx 1.17 on python 3.13 which doesn't work.
Once ONNX 1.18 is fully released and there are wheels for Python 3.13 available on PyPI, I'll delete my onnx2torch-py313 fork/published package and revert audio-separator to using the regular onnx2torch package.