microsoft

/

VibeVoice-ASR

like 386

Follow

Microsoft 17.9k

Thanks for sharing such a great model!
I've documented my process of running it with ROCm and Docker below.

1. Verify ROCm version using the following command: sudo update-alternatives --display rocm
2. Pull the PyTorch ROCm image select the appropriate image from rocm/pytorch
   Example used: rocm/pytorch:rocm7.1_ubuntu24.04_py3.12_pytorch_release_2.9.1
3. Start container, i use docker compose

services:
  pytorch-rocm:
    image: rocm/pytorch:rocm7.1_ubuntu24.04_py3.12_pytorch_release_2.9.1
    container_name: vibevoice-asr-rocm
    network_mode: "host"
    command: bash
    tty: true
    stdin_open: true
    devices:
      - /dev/kfd
      - /dev/dri
    group_add:
      - "video"
    ipc: host
    shm_size: "16gb"
    environment:
      TZ: "Asia/Taipei"
    mem_limit: "64g"
    restart: unless-stopped

4. Enter container and install inside container flash-attn rocm version

git clone https://github.com/ROCm/flash-attention.git &&\ 
cd flash-attention &&\
FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE" python setup.py install

5. Clone VibeVoice and install inside container
6. Run with gradio FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE" python demo/vibevoice_asr_gradio_demo.py --model_path microsoft/VibeVoice-ASR