FROM ghcr.io/ggml-org/llama.cpp:server

ENV PORT=7860
ENV LLAMA_CACHE=/tmp/llama-cache
RUN mkdir -p /tmp/llama-cache && chmod 777 /tmp/llama-cache

EXPOSE 7860

ENTRYPOINT ["/app/llama-server"]
CMD ["--hf-repo", "microsoft/Phi-3-mini-4k-instruct-gguf", \
     "--hf-file", "Phi-3-mini-4k-instruct-q4.gguf", \
     "--host", "0.0.0.0", \
     "--port", "7860", \
     "-c", "4096"]