Dockerfile

FROM debian:latest

RUN sudo apt update && sudo apt install -y aria2

WORKDIR /code
RUN git clone https://github.com/ggerganov/llama.cpp

WORKDIR /code/llama.cpp
RUN make LLAMA_OPENBLAS=1

RUN aria2 -d ./models/7B -o openhermes-2.5-mistral-7b.Q5_K_M.gguf -x 4 https://huggingface.co/TheBloke/OpenHermes-2.5-Mistral-7B-GGUF/resolve/main/openhermes-2.5-mistral-7b.Q5_K_M.gguf

CMD ["./server", "-m", "models/7B/openhermes-2.5-mistral-7b.Q5_K_M.gguf", "-t", "2", "-c", "4096", "--host", "0.0.0.0", "--port", "7860"]