Update Dockerfile

Dockerfile

@@ -25,9 +25,6 @@ RUN mkdir build && \
     cmake .. && \
     cmake --build . --config Release
 
-WORKDIR /data
-RUN wget https://huggingface.co/IlyaGusev/saiga2_70b_gguf/resolve/main/ggml-model-q4_1.gguf -nv -O model.gguf
-
 FROM ${BASE_CUDA_RUN_CONTAINER} as runtime
 
 
@@ -36,6 +33,11 @@ WORKDIR /app
 # Copy the executable from the build stage
 COPY --from=build /build/llama.cpp/build/bin/server /app
 COPY --from=build /data/model.gguf /data/model.gguf
+
+WORKDIR /data
+
+RUN wget https://huggingface.co/IlyaGusev/saiga2_70b_gguf/resolve/main/ggml-model-q4_1.gguf -nv -O model.gguf
+
 WORKDIR /app
 EXPOSE 7860