# 1. Use the highly optimized C++ server from the NEW official registry FROM ghcr.io/ggml-org/llama.cpp:server # 2. Expose the port Hugging Face expects for web traffic EXPOSE 7860 # 3. Download the model natively using Docker's ADD command. # This bypasses the need for apt-get entirely and assigns the correct permissions! ADD --chown=1000:1000 https://huggingface.co/bartowski/gemma-2-2b-it-GGUF/resolve/main/gemma-2-2b-it-Q4_K_M.gguf /models/gemma-2b.gguf # Switch to the non-root user required by Hugging Face Spaces USER 1000 # 4. Boot the server CMD ["-m", "/models/gemma-2b.gguf", "--host", "0.0.0.0", "--port", "7860", "-t", "2", "-c", "2048"]