Update Dockerfile

Dockerfile

@@ -18,7 +18,7 @@ ENV LLAMA_CUBLAS=0
 RUN mkdir build && \
     cd build && \
     cmake .. && \
-    cmake --build . --config Release
+    cmake --build . --config Release --target llama-server
 
 WORKDIR /data
 RUN wget https://huggingface.co/brunopio/Llama3-8B-1.58-100B-tokens-GGUF/resolve/main/Llama3-8B-1.58-100B-tokens-TQ2_0.gguf -nv -O model.gguf
@@ -28,7 +28,7 @@ FROM ${BASE_CPU_CONTAINER} as runtime
 WORKDIR /app
 
 # Copy the executable from the build stage
-COPY --from=build /build/llama.cpp/build/bin/server /app
+COPY --from=build /build/llama.cpp/build/bin/llama-server /app
 COPY --from=build /data/model.gguf /data/model.gguf
 COPY ./run.sh /app/run.sh
 WORKDIR /app