Dockerfile

# InferLoop Edge Runtime — CPU Experience Edition runtime image.
# Built from the closed-source artifacts produced by scripts/build.sh.
FROM ubuntu:24.04

RUN apt-get update \
    && apt-get install -y --no-install-recommends libgomp1 \
    && rm -rf /var/lib/apt/lists/*

WORKDIR /opt/inferloop-edge-runtime
COPY models/qwen/ ./models/qwen/
COPY artifacts/ ./_artifacts/
COPY conf/inferloop_edge.yaml ./conf/inferloop_edge.yaml
COPY scripts/run.sh ./run.sh
COPY scripts/check_oom_guard.sh ./check_oom_guard.sh
COPY scripts/collect_evidence.sh ./collect_evidence.sh

RUN mkdir -p ./lib ./bin ./plugin ./work \
    && cp ./_artifacts/libsdp.so ./lib/libsdp.so \
    && cp ./_artifacts/inferloop_edge ./bin/inferloop_edge \
    && cp ./_artifacts/libllm_agent_service.so ./plugin/ \
    && cp ./_artifacts/libllm_gpu_serving_service.so ./plugin/ \
    && cp ./_artifacts/libinferloop_agent_openai_service.so ./plugin/ \
    && cp ./_artifacts/libinferloop_static_web_service.so ./plugin/ \
    && cp ./_artifacts/test_inferloop_agent_services ./test_inferloop_agent_services \
    && for so in ./_artifacts/libllama.so* ./_artifacts/libggml*.so*; do \
         [ -e "$so" ] || continue; \
         cp -a "$so" ./lib/; \
       done \
    && for bin in test_llm_gpu_serving_cpu_ops test_llm_gpu_serving_qwen2_cpu; do \
         if [ -f "./_artifacts/$bin" ]; then \
           cp "./_artifacts/$bin" "./$bin"; \
           chmod 0555 "./$bin"; \
         fi; \
       done \
    && rm -rf ./_artifacts \
    && chmod 0555 ./run.sh ./check_oom_guard.sh ./collect_evidence.sh \
                  ./bin/inferloop_edge ./plugin/*.so \
                  ./test_inferloop_agent_services \
    && chmod 0755 ./work ./models ./models/qwen \
    && ldconfig -n ./lib

ENV LD_LIBRARY_PATH=/opt/inferloop-edge-runtime/lib:/opt/inferloop-edge-runtime/plugin
# Do not pin OMP_NUM_THREADS=1: the LLAMA backend uses llama.cpp n_threads /
# OpenMP workers. A process-wide OMP=1 collapses that pool. Prefer
# INFERLOOP_CPU_OP_THREADS (mapped to n_threads) as the parallelism knob.
ENV OPENBLAS_NUM_THREADS=1
ENV INFERLOOP_WORK_ROOT=/tmp/inferloop-edge-runtime
ENV INFERLOOP_MODEL_PROFILE=auto
ENV INFERLOOP_QWEN2_MODEL_DIR=/opt/inferloop-edge-runtime/models/qwen
ENV INFERLOOP_QWEN05B_MODEL=/opt/inferloop-edge-runtime/models/qwen
ENV INFERLOOP_CPU_OP_THREADS=8

ENTRYPOINT ["/opt/inferloop-edge-runtime/run.sh"]