M.I.M.I.R - Multi-agent Intelligent Memory & Insight Repository

# Build llama.cpp server for ARM64 (Apple Silicon) with Qwen2.5-VL # Supports vision-language models for image understanding FROM ubuntu:22.04 AS builder # Install build dependencies RUN apt-get update && apt-get install -y \ build-essential \ cmake \ git \ curl \ libcurl4-openssl-dev \ && rm -rf /var/lib/apt/lists/* # Clone llama.cpp repository WORKDIR /build RUN git clone https://github.com/ggerganov/llama.cpp.git . # Build llama.cpp server with multimodal support RUN mkdir build && cd build && \ cmake .. \ -DLLAMA_CURL=ON \ -DLLAMA_BUILD_SERVER=ON \ -DBUILD_SHARED_LIBS=OFF \ -DCMAKE_BUILD_TYPE=Release && \ cmake --build . --config Release -j$(nproc) # Runtime stage FROM ubuntu:22.04 # Install runtime dependencies RUN apt-get update && apt-get install -y \ libcurl4 \ curl \ libgomp1 \ && rm -rf /var/lib/apt/lists/* # Copy built binaries COPY --from=builder /build/build/bin/llama-server /usr/local/bin/llama-server # Create models directory RUN mkdir -p /models # Copy specific Qwen2.5-VL model (will be provided by build script) # Model size depends on variant: 2b (~1.5GB), 7b (~4.5GB), 72b (~45GB) ARG MODEL_FILE ARG VISION_FILE COPY docker/llama-cpp/models/${MODEL_FILE} /models/qwen2.5-vl.gguf COPY docker/llama-cpp/models/${VISION_FILE} /models/qwen2.5-vl-vision.gguf # Expose port EXPOSE 8080 # Health check HEALTHCHECK --interval=30s --timeout=10s --start-period=30s --retries=3 \ CMD curl -f http://localhost:8080/health || exit 1 # Set working directory WORKDIR /app # Default command with Qwen2.5-VL # Context size will be set by CMD override in build script # 2b: 32K tokens, 7b/72b: 128K tokens # --mmproj is required for vision capabilities ENTRYPOINT ["/usr/local/bin/llama-server"] CMD ["--host", "0.0.0.0", "--port", "8080", "--model", "/models/qwen2.5-vl.gguf", "--mmproj", "/models/qwen2.5-vl-vision.gguf", "--ctx-size", "131072", "--parallel", "4", "--threads", "-1"]