Dockerfile

# Base image
FROM nvidia/cuda:12.2.0-devel-ubuntu22.04

ENV DEBIAN_FRONTEND=noninteractive

# Update and install necessary dependencies
RUN apt update && \
    apt install --no-install-recommends -y \
        build-essential \
        python3 \
        python3-pip \
        wget \
        curl \
        git \
        cmake \
        zlib1g-dev \
        libblas-dev && \
    apt clean && \
    rm -rf /var/lib/apt/lists/*
    
# Setting up CUDA environment variables (this may not be necessary since you're using the official nvidia/cuda image, but it's good to be explicit)
ENV PATH="/usr/local/cuda/bin:$PATH" \
    LD_LIBRARY_PATH="/usr/local/cuda/lib64:$LD_LIBRARY_PATH" \
    CUDA_HOME="/usr/local/cuda"

WORKDIR /app

# Download ggml and mmproj models from HuggingFace
RUN wget https://huggingface.co/mys/ggml_bakllava-1/resolve/main/ggml-model-q4_k.gguf && \
    wget https://huggingface.co/mys/ggml_bakllava-1/resolve/main/mmproj-model-f16.gguf 

# Clone and build llava-server with CUDA support
RUN git clone https://github.com/ggerganov/llama.cpp.git && \
    cd llama.cpp && \
    git submodule init && \
    git submodule update && \
    make LLAMA_CUBLAS=1

# Create a non-root user for security reasons
RUN useradd -m -u 1000 user && \
    mkdir -p /home/user/app && \
    cp /app/ggml-model-q4_k.gguf /home/user/app && \
    cp /app/mmproj-model-f16.gguf /home/user/app

RUN chown user:user /home/user/app/ggml-model-q4_k.gguf && \
    chown user:user /home/user/app/mmproj-model-f16.gguf

USER user
ENV HOME=/home/user

WORKDIR $HOME/app

# Expose the port
EXPOSE 8080

# Start the llava-server with models
CMD ["/app/llama.cpp/server", "--model", "ggml-model-q4_k.gguf", "--mmproj", "mmproj-model-f16.gguf", "--threads", "4"]