Fix AMD rocm

[Alexander-Ger-Reich]

With this, it should work on AMD and NVIDIA graphics cards.

[zhu-han]

It seems that to support AMD, we only need to install the corresponding Torch version and avoid using torchaudio? In the recent update, I switched the audio I/O from torchaudio to soundfile+librosa. So the only remaining task is to add instructions for installing the correct Torch version. If you find this is the case, maybe you could open a PR to add the Torch installation instructions under “Step 1: Install PyTorch”.

[Alexander-Ger-Reich]

ya. torchaudio has problems. https://rocm.docs.amd.com/projects/radeon-ryzen/en/latest/docs/install/installrad/windows/install-pytorch.html

pip install --upgrade https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_core-7.2.1-py3-none-win_amd64.whl https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_libraries_custom-7.2.1-py3-none-win_amd64.whl https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm-7.2.1.tar.gz https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torch-2.9.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torchaudio-2.9.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torchvision-0.24.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl

[bluemoehre]

I am using the image provided by ROCm / AMD:

FROM docker.io/rocm/pytorch:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0

WORKDIR /app

# Install flash attention with triton backend
RUN FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE pip install flash-attention --no-build-isolation

# System deps
RUN apt-get update && apt-get install -y \
    git \
    ffmpeg \
    && rm -rf /var/lib/apt/lists/*

# Python deps - choose one
#RUN pip install --no-cache-dir omnivoice
RUN pip install git+https://github.com/k2-fsa/OmniVoice.git

# Startup
CMD ["omnivoice-demo", "--ip", "0.0.0.0", "--port", "8880"]

[EnochPrime]

You can avoid needed users to uncomment lines by using the optional dependencies methods like in #143.