Upgrade to the upstream fixes

Dockerfile

@@ -1,13 +1,18 @@
-FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
+FROM nvidia/cuda:12.1.1-devel-ubuntu22.04
 ARG MODEL
+ARG IMGMODEL
+ARG WHISPERMODEL
+ARG MMPROJ
 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
+RUN apt update && apt install curl -y
 WORKDIR /opt/koboldcpp
 COPY default.json /opt/koboldcpp/default.json
-RUN make LLAMA_OPENBLAS=1 LLAMA_CUBLAS=1 LLAMA_PORTABLE=1
-RUN wget -O model.ggml $MODEL
-CMD /bin/python3 ./koboldcpp.py --model model.ggml $ADDITIONAL --port 7860 --hordeconfig $MODEL_NAME 1 1 --preloadstory default.json
-
+RUN curl -fLO koboldcpp https://koboldai.org/cpplinuxcu12
+RUN chmod +x ./koboldcpp
+RUN curl -fLO model.ggml $MODEL || true
+RUN curl -fLO imgmodel.ggml $IMGMODEL || true
+RUN curl -fLO mmproj.ggml $MMPROJ || true
+RUN curl -fLO whispermodel.ggml $WHISPERMODEL || true
+CMD ./koboldcpp --model model.ggml --whispermodel whispermodel.ggml --sdmodel imgmodel.ggml --sdthreads 4 --sdquant --sdclamped --mmproj mmproj.ggml $ADDITIONAL --port 7860 --hordemodelname $MODEL_NAME --hordemaxctx 1 --hordegenlen 1 --preloadstory default.json --ignoremissing