Dockerfile

FROM nvidia/cuda:12.1.1-cudnn8-runtime-ubuntu22.04

ENV DEBIAN_FRONTEND=noninteractive
ENV PATH="/root/.local/bin:$PATH"

# Fix for getpwuid() error when no user is present
RUN echo "user:x:1000:1000::/home/user:/bin/bash" >> /etc/passwd && \
    mkdir -p /home/user && chown 1000:1000 /home/user

# Install dependencies
RUN apt-get update && apt-get install -y \
    python3.10 \
    python3.10-venv \
    python3.10-distutils \
    python3-pip \
    curl \
    git \
    ffmpeg \
    libsndfile1 \
    && apt-get clean

# Set python and pip aliases
RUN ln -sf /usr/bin/python3.10 /usr/bin/python && \
    ln -sf /usr/bin/pip3 /usr/bin/pip

# Install uv and latest vllm[audio] from nightly index
RUN curl -Ls https://astral.sh/uv/install.sh | bash && \
    ~/.local/bin/uv pip install --system "vllm[audio]" --extra-index-url https://wheels.vllm.ai/nightly

# Install Hugging Face tools
RUN pip install huggingface_hub

# Preload model weights (optional but recommended)
RUN python -c "from huggingface_hub import snapshot_download; snapshot_download('mistralai/Voxtral-Mini-3B-2507')"

# Expose the OpenAI-compatible API port
EXPOSE 8000

# Final run command (corrected)
CMD ["vllm", "serve", "mistralai/Voxtral-Mini-3B-2507", "--tokenizer_mode", "mistral", "--config_format", "mistral", "--load_format", "mistral"]