FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
RUN apt update && apt install git build-essential libopenblas-dev wget python3-pip -y
RUN apt update && apt install -y wget
ARG MODEL_PART_A
ARG MODEL_PART_B
ARG MODEL_NAME
ARG ADDITIONAL
RUN mkdir /opt/koboldcpp
RUN apt update && apt install git build-essential libopenblas-dev wget python3-pip -y
RUN git clone https://github.com/lostruins/koboldcpp /opt/koboldcpp
WORKDIR /opt/koboldcpp
COPY default.json /opt/koboldcpp/default.json
RUN make LLAMA_OPENBLAS=1 LLAMA_CUBLAS=1 LLAMA_PORTABLE=1
RUN wget -O part_a.gguf $MODEL_PART_A
RUN wget -O part_b.gguf $MODEL_PART_B
RUN cat part_a.gguf part_b.gguf > model.gguf
RUN rm part_a.gguf part_b.gguf
CMD /bin/python3 ./koboldcpp.py --model model.gguf $ADDITIONAL --port 7860 --hordeconfig $MODEL_NAME 1 1 --preloadstory default.json