pymss

Python package for music source separation.
[English] 简体中文

Install

pymss 2.2 and later requires Python 3.11 or newer.

If you want the CUDA build of PyTorch, install it first:

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128

Intel XPU (Arc)

Linux requires working Level Zero and OpenCL drivers. Install the official PyTorch XPU wheel before pymss:

pip install torch --index-url https://download.pytorch.org/whl/xpu
pip install pymss

Use --device xpu to force Intel GPU execution and --device-id to select a device. When multiple IDs are supplied, only the first is used; multi-XPU parallel execution is not supported. --device auto prefers CUDA, then XPU, NPU, and Apple Silicon. XPU hardware inference is experimental and explicitly uses FP32, even when use_amp=True. bf16 AMP requires separate model accuracy and performance validation.

A manual benchmark script is available at test/manual/xpu_benchmark.py to compare FP32 CPU and XPU inference performance on local hardware.

AMD ROCm

pip install torch --index-url…