matthoffner/ggml-coding-llm

1FROM ghcr.io/ggerganov/llama.cpp:full-cuda2 3ENV DEBIAN_FRONTEND=noninteractive4 5# Update and install necessary dependencies6RUN apt update && \7    apt install --no-install-recommends -y \8        build-essential \9        python3 \10        python3-pip \11        wget \12        curl \13        git \14        cmake \15        zlib1g-dev \16        libblas-dev && \17    apt clean && \18    rm -rf /var/lib/apt/lists/*19 20# 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)21ENV PATH="/usr/local/cuda/bin:$PATH" \22    LD_LIBRARY_PATH="/usr/local/cuda/lib64:$LD_LIBRARY_PATH" \23    CUDA_HOME="/usr/local/cuda"24 25WORKDIR /app26 27RUN wget https://huggingface.co/itsdotscience/Magicoder-S-DS-6.7B-GGUF/resolve/main/Magicoder-S-DS-6.7B_q8_0.gguf28 29RUN make LLAMA_CUBLAS=130 31# Expose the port32EXPOSE 808033 34# Start the llava-server with models35CMD ["--server", "--model", "Magicoder-S-DS-6.7B_q8_0.gguf", "--threads", "6", "--host", "0.0.0.0", "-ngl", "33"]36