Dockerfile

FROM nvidia/cuda:11.8.0-devel-ubuntu22.04

# Install necessary packages
RUN apt update && apt install -y \
    git \
    build-essential \
    libopenblas-dev \
    wget \
    python3-pip \
    nodejs \
    npm

# Create a new user to avoid using root
RUN useradd -m -u 1000 user

# Switch to the new user
USER user

# Set environment variables
ENV HOME=/home/user \
    PATH=/home/user/.local/bin:$PATH

# Set the working directory
WORKDIR $HOME/app

# Copy the current directory contents into the container at /home/user/app
COPY --chown=user . $HOME/app

# Install aphrodite-engine from PyPI to handle dependencies
RUN python3 -m pip install aphrodite-engine

# Clone the specific branch of aphrodite-engine for the latest features
RUN git clone --branch feat/exllamav2-support https://github.com/PygmalionAI/aphrodite-engine.git $HOME/aphrodite-engine

# Install additional dependencies
RUN pip install huggingface-hub hf-transfer

# Set environment variable to enable hf-transfer
ENV HF_HUB_ENABLE_HF_TRANSFER=1

# Download the model using huggingface-cli
RUN huggingface-cli download LoneStriker/TinyLlama-1.1B-32k-Instruct-8.0bpw-h8-exl2 --local-dir $HOME/goliath-gptq --local-dir-use-symlinks False --cache-dir $HOME/cache

# Expose the port the API server will listen on
EXPOSE 7860

# Command to run the API server from the cloned directory
CMD ["/bin/bash", "-c", "cd $HOME/aphrodite-engine && /bin/python3 -m aphrodite.endpoints.openai.api_server -q exl2 --dtype auto -gmu 0.95 --kv-cache-dtype fp8_e5m2 --max-num-seqs 15 --served-model-name "BagelMIsteryTour-v2-8x7B-AWQ" --enforce-eager -tp 4 --port 7860 --host 0.0.0.0 --model ~/goliath-gptq"]