Prepare Docker file ahead of multi-backend builds of llama-server

llama.cpp can be built to support multiple variants of the CPU backend,
and choose the best one at runtime. This requires a dynamically linked
build of the llama-server. So modify the model-runner Docker file to
play nice with such builds of the llama-server.

Signed-off-by: Piotr Stankiewicz <[email protected]>

Author: Piotr Stankiewicz

Dockerfile

@@ -53,7 +53,7 @@ COPY --from=builder /app/model-runner /app/model-runner
 
 # Copy the llama.cpp binary from the llama-server stage
 ARG LLAMA_BINARY_PATH
-COPY --from=llama-server ${LLAMA_BINARY_PATH}/bin/com.docker.llama-server /app/bin/com.docker.llama-server
+COPY --from=llama-server ${LLAMA_BINARY_PATH}/ /app/.
 RUN chmod +x /app/bin/com.docker.llama-server
 
 USER modelrunner
@@ -64,6 +64,7 @@ ENV MODEL_RUNNER_PORT=12434
 ENV LLAMA_SERVER_PATH=/app/bin
 ENV HOME=/home/modelrunner
 ENV MODELS_PATH=/models
+ENV LD_LIBRARY_PATH=/app/lib
 
 # Label the image so that it's hidden on cloud engines.
 LABEL com.docker.desktop.service="model-runner"